Automated mortality risk assessment through blood smear images and subject data

Abstract

Presented herein are systems and systems of determining values indicating probabilities of conditions of subjects using blood samples. A computing system may receive a plurality of images of a blood sample having white blood cells (WBCs) and red blood cells (RBCs) obtained from a subject at a time. The computing system may identify, from the plurality of images of the blood sample, (i) a set of WBC images corresponding to the WBCs and (ii) a set of RBC images corresponding to the RBCs. The computing system may apply the set of WBC images and the set of RBC images to a machine learning (ML) architecture. The computing system may determine, based on applying the ML architecture, a value indicating a probability of mortality for the subject at the time interval relative to the time. The computing system may generate a classification of the first subject in accordance with the value.

Description

Atty. Dkt. No.: 115872-3344AUTOMATED MORTALITY RISK ASSESSMENT THROUGH BLOOD SMEAR IMAGES AND SUBJECT DATACROSS REFERENCE TO RELATED APPLCIATIONS

[0001] The present application claims the benefit of and priority to U.S. ProvisionalPatent Application No. 63 / 715,856, filed November 4, 2024, which is incorporated herein by reference in its entirety.BACKGROUND[0002| A computing device may use a machine learning model to process an input to generate an output.SUMMARY

[0003] Aspects of the present disclosure are directed to systems and methods of determining values indicating probabilities of conditions of subjects using blood samples. One or more processors may receive a first plurality of images of a first blood sample having first white blood cells (WBCs) and first red blood cells (RBCs) obtained from a first subject at a first time. The one or more processors may identify, from the first plurality of images of the first blood sample, (i) a first set of WBC images corresponding to the first WBCs and (ii) a first set of RBC images corresponding to the first RBCs. The one or more processors may apply the first set of WBC images and the first set of RBC images to a machine learning (ML) architecture. The ML architecture can be established using a plurality of examples. Each of the plurality of examples can include (i) a second set of WBC images of a second blood sample from a second subject at a second time, (ii) a second set of RBC images of the second blood sample from the second subject at the second time, and (iii) a label indicating one of mortality or survival at a time interval relative to the second time. The one or more processors may determine, based on applying the first plurality of images to the ML architecture, a value indicating a probability of mortality for the first subject at the time interval relative to the first time. The one or moreAtty. Dkt. No.: 115872-3344 processors may generate a classification of the first subject as one of mortality or survival in accordance with the value indicating the probability of mortality. The one or more processors may may store, using one or more data structures, an association between the first subject and the classification.

[0004] In some embodiments, the one or more processors may generate to indicate that the first subject is to survive for the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold. In some embodiments, the one or more processors may provide an output identifying the classification to indicate that the first subject is to survive for the time interval relative to the first time. In some embodiments, the one or more processors may generate the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold. In some embodiments, the one or more processors may provide an output identifying the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time.

[0005] In some embodiments, the one or more processors may provide, based on the classification, the output and can include at least one of: (i) a notification for a clinician to examine the first subject, (ii) a notification to administer an intervention within the time interval, or (iii) a notification for the first subject to request for medical attention. In some embodiments, the one or more processors may receive a first non-image dataset that can include at least one of (i) a first plurality of traits of the first subject, (ii) a first blood count derived from the first blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject. In some embodiments, the method can include applying to the ML architecture further and can include applying the first non-image dataset to the ML architecture. At least one of the plurality of examples can include a second non-image dataset and can include at least one of (i) a second plurality of traits of the second subject, (ii) a second blood count derived from the second blood sample, (iii) a second plurality of measures derived from testing of the second blood sample, or (iv) a second plurality of physiological measurements of the second subject. In some embodiments, the one or moreAtty. Dkt. No.: 115872-3344 processors may generate classification based on applying the first non-image dataset to the ML architecture.

[0006] In some embodiments, at least one of the plurality of examples further can include the label identifying one of a presence or absence at least one of a plurality of conditions associated with the mortality in the second subject, wherein the plurality of conditions can include a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MAHA), acidosis, multi-organ failure, anemia, hemophagocytic lymphhistiocytosis, and cytokine release syndrome. In some embodiments, the one or more processors can determine based on applying the first plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the first subject. In some embodiments, the one or more processors may generate the classification to identify one of a presence or absence of a condition of the plurality of conditions associated with the mortality in the first subject in accordance with the second value. In some embodiments, the one or more processors can generate based on applying the first plurality of images to the ML architecture, a plurality of embeddings used to determine the second value. In some embodiments, the one or more processors can execute using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space. Each of the plurality of clusters can be associated with a respective set of parameters for at least one of the plurality of conditions. In some embodiments, the one or more processors can determine, based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters. In some embodiments, the one or more processors can identify from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.[0 07| In some embodiments, at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from testing of the second blood sample. The clustering model is established by determining, for each cluster of the plurality of clusters, the respective set of parameters based on the plurality of parameters of the at least oneAtty. Dkt. No.: 115872-3344 example associated with a second plurality of embeddings assigned to the cluster. In some embodiments the one or more processors can determine a relative score indicating a difference between the first subject and a plurality of subjects based on the value indicating the probability of mortality for the first subject and a second value indicating a composite probability of mortality of a plurality of subjects. In some embodiments, the one or more processors can identify, from a plurality of categories, a category for the first subject in accordance with the relative score and a score category for the category.

[0008] In some embodiments, the one or more processors can provide for presentation, a user interface comprising one or more of: (i) a category for the first subject, (ii) a relative score between the first subject and a plurality of subjects, (iii) the value indicating the probability of mortality for the first subject, (iv) the set of parameters for a condition, (v) the condition of the first subject. In some embodiments, the first blood sample is acquired in accordance with peripheral blood smear (PBS) and the first plurality of images of the first blood sample is generated within a predefined time of the PBS. In some embodiments, the one or more processors may identify at least one image from the first plurality of images as one of the first set of WBC images and the first set of RBC images based on a visual characteristic of the at least one image.

[0009] In some embodiments, the ML architecture can include a WBC encoder configured to generate a set of WBC embeddings using the first set of WBC images, an RBC encoder configured to generate a set of RBC embeddings using the first set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the first subject based on the set of WBC embeddings, the set of RBC embeddings, and the set of feature embeddings, and a classifier configured to generate the classification using the value and a non-image dataset. In some embodiments, the time interval can include at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 60 days, 90 days, 120 days, or 180 days.Atty. Dkt. No.: 115872-3344[00101 In some embodiments, the first subject is at risk of or diagnosed with cancer, wherein the cancer can include at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.(0011 J Aspects of the present disclosure are directed to systems and methods of training a machine learning model to determine values indicating probabilities of conditions of subjects using blood samples. One or more processors may retrieve a training dataset including a plurality of examples. Each of the plurality of examples can include: (i) a plurality of images of a blood sample having white blood cells (WBCs) and red blood cells (RBCs) obtained from a subject at a time and (ii) a label indicating one of mortality or survival at a time interval relative to the time. The one or more processors may identify, from the plurality of images of the blood sample of at least one example of the plurality of examples in the training dataset, (i) a set of WBC images corresponding to the WBCs and (ii) a set of RBC images corresponding to the RBCs. The one or more processors may apply the set of WBC images and the set of RBC images to a machine learning (ML) architecture and can include a plurality of weights to determine a value indicating a probability of mortality for the subject at the time interval relative to the time. The one or more processors may generate a classification of the subject as one of mortality or survival in accordance with the value indicating the probability of mortality. The one or more processors may compare the classification of the subject generated in accordance with the value determined by the ML architecture with the label of the training dataset. The oneAtty. Dkt. No.: 115872-3344 or more processors may update at least one of the plurality of weights in the ML architecture based on comparing the classification and the label.

[0012] In some embodiments, at least one of the plurality examples further can include a non-image dataset including at least one of (i) a plurality of traits of the subject, (ii) a blood count derived from the blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject. In some embodiments, the one or more processors may apply the non-image dataset to the ML architecture, and generate the classification based on applying the non-image dataset to the ML architecture. In some embodiments, at least one of the plurality examples can include the label identifying a presence or an absence of at least one of a plurality of conditions associated with the mortality in the second subject. In some embodiments, the plurality of conditions can include a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (D1C), microangiopathic hemolytic anemia (MHA), acidosis, multi-organ failure, anemia hemophagocytic lymphhistiocytosis, and cytokine release syndrome. In some embodiments, the one or more processors may generate the classification to identify a condition of the plurality of conditions associated with the mortality in the subject. In some embodiments, the one or more processors can determine, based on applying the plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the first subject. In some embodiments the one or more processors can generate the classification to identify a presence or an absence of a condition of the plurality of conditions associated with the mortality in the subject in accordance with the second value. In some embodiments, the one or more processors may update at least one of the plurality of weights based on comparing the condition identified by the classification and the label.[00131 In some embodiments, the one or more processors can generate, based on applying the first plurality of images to the ML architecture, a plurality of embeddings used to determine the second value. The one or more processors can execute, using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space, each ofAtty. Dkt. No.: 115872-3344 the plurality of clusters associated with a respective set of parameters for at least one of the plurality of conditions. The one or more processors can determine, based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters. The one or more processors can identify, from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.[0014 In some embodiments, at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from testing of the second blood sample. The clustering model is established by determining, for each cluster of the plurality of clusters, the respective set of parameters based on the plurality of parameters of the at least one example associated with a second plurality of embeddings assigned to the cluster. The one or more processors can determine a relative score indicating a difference between the first subject and a plurality of subjects based on the value indicating the probability of mortality for the first subject and a second value indicating a composite probability of mortality of a plurality of subjects. The one or more processors cam identify, from a plurality of categories, a category for the first subject in accordance with the relative score and a score category for the category.

[0015] In some embodiments, the blood sample is acquired in accordance with peripheral blood smear (PBS). In some embodiments, the one or more processors may identify at least one image from the plurality of images as one of the sets of WBC images and the set of RBC images based on a visual characteristic of the at least one image. In some embodiments, the ML architecture can include a WBC encoder configured to generate a set of WBC embeddings using the set of WBC images, an RBC encoder configured to generate a set of RBC embeddings using the set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the subject based on the set of WBC embeddings, the set of RBC embeddings, the set of feature embeddings, and a classifier configured to generate the classification using the value and a non-image dataset.Atty. Dkt. No.: 115872-3344[0016| In some embodiments, the time interval can include at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 60 days, 90 days, 120 days, or 180 days. In some embodiments, the subject is at risk of or diagnosed with cancer, wherein the cancer can include at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.BRIEF DESCRIPTION OF THE DRAWINGS[0017| The foregoing and other objects, aspects, features, and advantages of the disclosure will become more apparent and better understood by referring to the following description taken in conjunction with the accompanying drawings, in which:

[0018] FIG. 1 : Example of a machine learning (ML) architecture.

[0019] FIGs. 2A-2B: Example results from the use of the ML architecture.

[0020] FIGs. 3A-3B: Example results based on varying images tabular data and probabilities output from the ML architecture.

[0021] FIG. 4: Examples of correct classifications of red blood cells (RBC) and white blood cells (WBC) impacted by various conditions in deceased patients.Atty. Dkt. No.: 115872-3344[00221 FIG. 5: Examples of correct classifications red blood cells (RBC) and white blood cells (WBC) impacted by various conditions in controlled patients.

[0023] FIG. 6A-6B: Examples of correct classifications based on blood smears of subjects.

[0024] FIG. 7A-7B: Examples of classifiers for cell morphology.[00251 FIG. 8A-8B: Examples of learning curves for the ML architecture.

[0026] FIG. 9A-9B: Examples of correct classifications from the ML architecture based on the actual classifications of the of the WBC and the RBC.

[0027] FIG. 10: An example table indicating a plurality of conditions, a number of cases for each condition, a risk of mortality for each condition, how often the condition is detected by a physician.

[0028] FIG. 11A-C: Example of cells that are afflicted with a condition that corresponds to a mortality of the subject.

[0029] FIG. 12: A graph indicating a performance of a machine learning model executing per condition.

[0030] FIG. 13A and 13B: Graphs of relative risk of death and probability of survival across risk groups and patient cohort groups.

[0031] FIG. 14: A table listing model performance metrics across different patient risk groups in both test and control trial validation setting.

[0032] FIG. 15: A graph of log odd ratios for specific red blood cell (RBC) or white blood cell (WBC) morphologies for mortality fingerprint.Atty. Dkt. No.: 115872-3344[00331 FIG. 16: A graph of log odd ratios for specific red blood cell (RBC) or white blood cell (WBC) morphologies for early detection of myelodysplastic syndromes (MDS) from peripheral blood.|0034[ FIGs. 17A-C: Graphs of clustering of patient subgroups based on peripheral blood smear morphology, specifically focusing on echinocyte prevalence and its biochemical correlates.

[0035] FIG. 18A: A set of box plots comparing a range of morphological and laboratory features between deceased patients and controls. Each plot shows the distribution of a specific feature in both groups, along with the percentage of deceased patients exhibiting abnormal values.| 0036] FIG. 18B: Another set of box plots comparing a range of morphological and laboratory features between deceased patients and controls. Each plot shows distributions of a specific feature, along with the percentage of deceased patients demonstrating abnormal values.[0037| FIG. 19: A graph of ranking of predictive importance of individual features for mortality patients % outside 95% confidence interval versus control.

[0038] FIG. 20: A graph of ranking of predictive importance of individual features based on the Interquartile Range (IQR) Outlier method.

[0039] FIG. 21 is a block diagram of a system of determining values indicating probabilities of mortality of subjects using blood samples, in accordance with an illustrative embodiment.

[0040] FIG. 22A is a block diagram of a process of training a machine learning architecture (ML) to determine values indicated an expected mortality of subjects using blood samples, in accordance with an illustrative embodiment.Atty. Dkt. No.: 115872-3344[00411 FIG. 22B is a block diagram of a process of training a clustering model to determine, for each cluster, a set of parameters, in accordance with an illustrative embodiment.

[0042] FIG. 23 A is a block diagram of a process of generating a classification indicating a presence or absence of conditions indicating mortality of subjects using blood samples, in accordance with an illustrative embodiment.

[0043] FIG. 23B is a block diagram of a process of determining an assignment using a clustering model, in accordance with an illustrative embodiment.[0044| FIG. 24 is a block diagram of a process of providing an output to an administrative device, in accordance with an illustrative embodiment.[0045| FIG. 25 is an example of a user interface displayed on the administrative device.[0046| FIG. 26 is a flow diagram of a method of determining values indicating probabilities of mortality of subjects using blood samples, in accordance with an illustrative embodiment.[0047J FIG. 27 is a flow diagram of method of training machine learning (ML) architectures to determine values indicating expected mortality of subjects using blood samples, in accordance with an illustrative embodiment.[0048| FIG. 28 is a block diagram of a computing environment according to an example implementation of the present disclosure.DETAILED DESCRIPTION[0049 J Following below are more detailed descriptions of various concepts related to, and embodiments of, systems and methods for generating determining values indicating probabilities of conditions of subjects using blood samples. It should be appreciated that various concepts introduced above and discussed in greater detail below may be implemented in any of numerous ways, as the disclosed concepts are not limited to any particular manner of implementation.Atty. Dkt. No.: 115872-3344Examples of specific implementations and applications are provided primarily for illustrative purposes.

[0050] Section A describes the Deep Learning-Driven Mortality Risk Assessment throughPeripheral Blood Smear Morphology in Hospitalized Cancer Patients.10051] Section B describes systems and methods for determining values indicating probabilities of condition of subjects using blood samples.

[0052] Section C describes a network and computing environment.A. Deep Learning-Driven Mortality Risk Assessment through Peripheral Blood Smear Morphology in Hospitalized Cancer Patients

[0053] Introduction: Mortality in cancer patients is often preceded by catastrophic hematologic and inflammatory syndromes such as the systemic inflammatory response syndrom e / sepsis, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia, hemophagocytic lymphohistiocytosis, and cytokine release syndrome. These conditions alter peripheral blood cells, causing changes in the morphology of red blood cells (RBC), white blood cells (WBC) and platelets in peripheral blood smears (PBS). Attention-based deep learning architectures was hypothesized that using PBS feature embeddings can predict short-term mortality risk and quantify the underlying biology contributing to that risk.

[0054] Methods: 790,000 PBS was utilized, with corrected cell level clinical annotations. This dataset included 790,000 RBC / platelet and 100 million WBC single cell morphology images. 20,000 RBC / platelet and 20,000 WBC images were selected to train two ResNext50 architectures that generate feature embeddings. The cohort consisted of 627 PBS from patients who died within 24 hours and 15,863 control PBS, split 60 / 20 / 20% respectively for training, validation, and testing. These slides came from 1,565 unique patients with no overlap between the test and training and validation sets. Features were extracted and trained a multiple instance deep neural network withAtty. Dkt. No.: 115872-3344 self-attention to predict 24-hour mortality risk. A logistic regression classifier using 8 CBC parameters and patient age served as the baseline architecture.

[0055] The top 10% of highest attention cells were analyzed from correctly predicted mortality events (mortality-wbc, mortality-rbc) and control cases (control-wbc, control-rbc) to identify morphologies linked to increased mortality risk. A pathologist performed a qualitative analysis, and the two ResNext classifiers were used to perform a quantitative analysis of class distribution changes between the two sets using a two-proportion z-test.

[0056] Results[0057| Performance: When run on a test set with 3,172 slides from control patients, and 134 slides from deceased patients, the trained MIL architecture achieved an AUC = 0.83. The logistic regression architecture using CBC and age had an AUC = 0.71.10058] Attention-Based Pathologist Morphology Analysis :

[0059] mortality-rbc. “anisopoikilocytosis with increased echinocytes, schistocytes, tear drops, and spherocytes, increased polychromatophilic cells.

[0060] control-rbc Unremarkable.

[0061] mortality-wbc. Increased erythroblasts, often with abnormal / dysplastic nuclei, increased large / giant platelets, increased smudge cells, some of which have neutrophil extracellular trap (NET) morphology.”[0062| control-WBC'. “Normal neutrophils, lymphocytes or monocytes.”

[0063] Computational Quantitative Analysis:[0064| When comparing the top 10% highest attention RBC patches, the mortality group showed an increase in echinocytes (54%:0%) and schistocytes (29%:3%), and a decrease in normal RBC patches (4%:72%) (all p<10-10). For WBCs, the mortality group had more erythroblastsAtty. Dkt. No.: 115872-3344(44%: 1%), smudge cells (14%: 1%), and giant thrombocytes (12%: 1%), with fewer segmented neutrophils (5%:44%), band neutrophils (6%: 15%), and lymphocytes (4%: 13%) (all p<10-10).

[0065] Conclusions: Deep learning-based analysis of PBSs could provide clinical decision support in hospitalized patients by assessing their 24-hour mortality risk. These patients may require escalation of care and specific interventions depending on the nature of the risk. Qualitative and quantitative analyses of the architecture’s attention scores suggest echinocytes, erythroblastosis, schistocytes, large platelets, smudge cells and NETs were heavily relied on. The elevation of erythroblasts, giant platelets, and schistocytes may signal a profound marrow response to severe hemolysis, tissue hypoxia, and consumptive coagulopathy as seen in severe DIC.

[0066] FIG. 1 : Example of a machine learning (ML) architecture. The architecture for this model can be a multi-encoder, multi-model, multiple instance model (ML) 100. FIG. 1 describes the architecture of the ML model used to make mortality predictions. The inputs can be white blood cell (WBC) images and red blood cell (RBC patches), each with their own image encoder (multi-encoder). Images are encoded using a specially trained convolutional neural network described in “Cell Morphology Classifiers” section of this presentation. The architecture can further be modified to intake additional sets of images, such as individual RBCs or platelets. In addition, the encoders can be swapped out for further customization.[0067J Embeddings are then run through a multiple instance learning (ML) network and output a mortality probability. That probability is then combined with age and parameters from the complete blood count with a random forest to make a final multimodal mortality prediction, the percent change of death for the given time period. Additional multimodal data can be inserted at this point, such as additional laboratory data or other information that influences patient risk, such as their service or location in the hospital. The ML model can include an attention layer that is used to apply attention scores to the individual images, which enables the model to highlight cells that are most contributing to the risk, which is helpful for clinicians to interpret the cause of the mortality risk.Atty. Dkt. No.: 115872-3344[0068| FIGs. 2A-2B: Example results from the use of the ML architecture. FIG. 2A depicts the Area Under the Receiver Operator Curve (AUC) for a version of the model trained to predict mortality risk within 24 hours using both image data and tabular data. Since the model’s cutoff can be changed to optimize for different parameters such as sensitivity, specificity, positive or negative predictive value, AUC is used to measure overall performance. FIG. 2B an optimized model cutoff value was chosen. 1 indicates death and 0 indicates no death. For this cutoff of the ML model, it has the following performance at predicting 24-hour mortality events. It correctly detects 78% of all deaths and 95% of all nondeaths. It is important to note, the algorithm can be used in this way to make a binary classification, or the result of the algorithm can be provided as a death risk score, which is the percentage change of death within the given time frame.[0069| FIGs. 3A-3B: Example results based on varying images tabular data and probabilities output from the ML architecture. FIG. 3A is the Area Under the Receiver Operator Curve (AUC) for different versions of the model that make predictions within different timeframes. Since the model’s cutoff can be changed to optimize for different parameters such as sensitivity, specificity, positive or negative predictive value, AUC is used to measure overall performance. FIG. 3A shows that image + tabular data outperforms either image data alone or tabular data alone. FIG. 3B, an optimized model cutoff is used, and the results are analyzed in predicting death within 5 days using an optimized cutoff. 1 indicates death and 0 indicates no death. 83% of deaths were captured correctly by the model. 98% of nondeaths were correctly predicted by the model. It is important to note, the algorithm can be used in this way to make a binary classification, or the result of the algorithm can be provided as a death risk score, which is the percentage change of death within the given time frame.

[0070] FIG. 4: Examples of correct classifications of red blood cells (RBC) and white blood cells (WBC) impacted by various conditions in deceased patients. This is an attention signature for a patient who died within 24 hours. The top images are those that have most lead the algorithm to predict that the patient will die. In the RBC compartment, they show the patient has schistocytes, which are red blood cell fragments that can be seen in life threatening conditions such as disseminated intravascular coagulation (DIC) and microangiopathic hemolytic anemiaAtty. Dkt. No.: 115872-3344(MAHA). The second row shows the white blood cells contributing to the death signature, which show many abnormal blast cells, which strongly suggest the patient has acute myeloid or acute lymphoid leukemia.|0071 J The bottom row is a set of randomly selected control images for this same patient, to show how the attention signature has picked up the abnormal cells. It is seen that only 20% of RBC patches are schistocytes for this patient and that the blast percentage is quite low. In this case, the Al has correctly extracted the pathologic cell morphologies from the input data and placed them in this disease signature graph.

[0072] FIG. 5: Examples of correct classifications red blood cells (RBC) and white blood cells (WBC) impacted by various conditions in controlled patients. FIG. 5 depicts an attention signature for a patient who did not die. The red blood cells are mostly normal, with a few teardrops, which are not life threatening. The white blood cells are also a normal assortment of lymphocytes and neutrophils, the most common cells found in normal circulation.

[0073] FIG. 6A-6B: Examples of correct classifications based on blood smears of subjects. This graph analyzes the difference between high attention cells in patients who died versus those who did not die. It shows that for patients who die, that algorithm is identifying and paying extra attention to blasts (a sign of acute leukemia), band neutrophils, a sign of acute bacterial infection, and erythroblasts, a sign of marrow stress and blood loss. On the RBC side, it is seen that patients who die are more likely to have schistocytes, which are RBC fragments that are created during microangiopathic hemolytic anemia and disseminated intravascular coagulation, both of which can be seen in near death patients. Increased echinocytes can be seen, which can be seen in liver failure, kidney failure, or may be the result of blood acidosis. Increased macrocytes can be seen, which can be seen in response to bleeding. Non-death samples had a large proportion of lymphocytes, monocytes and segmented neutrophils, all of which are normally found in circulation. Normal morphology was also the most common RBC morphology in the non-death patients. These findings underscore that the biological mechanism behind which the algorithm is making its inferences.Atty. Dkt. No.: 115872-3344[0074| FIG. 7A-7B: Examples of classifiers for cell morphology. The above figure shows how the data is generated to train the RBC patch image classifier. A dataset was used of over 500,000 patient slides. From those, the top 1% was identified and top 10% of images witch each morphology per the medical record. That label was then assigned to RBC region image patches derived from that sample. These images were then trained using a convolutional neural network to make predictions.[0075[ FIG. 8A-8B: Examples of learning curves for the ML architecture. The graph within FIG. 8A shows the accuracy of the RBC and WBC classifiers over the training epochs on both the training and validation sets. The graph within FIG. 8B shows the training loss during this same time frame.

[0076] FIGs. 9A-9B: Examples of correct classifications from the ML architecture based on the actual classifications of the of the WBC and the RBC. Feature embeddings for the M3M1L architecture are generated by convolutional neural networks that analyze WBC and RBC morphology. The confusion matrixes in FIGs. 9A-9B show the performance of these algorithms at cell classification.

[0077] FIG. 10: depicts an example of a table indicating a plurality of conditions, a number of cases for each condition, a risk of mortality for each condition, how often the condition is detected by a physician, and opportunities for artificial intelligence (Al)-driven diagnostics. The table can include syndromes such as sepsis, septic shock, hospital-acquired infection, neutropenic sepsis, multi-organ failure (MOF), disseminated intravascular coagulation (DIC), acidosis, hemophagocytic lymphohistiocytosis (HLH), and anemia, alongside estimated U.S. annual case counts, mortality percentages, and total annual healthcare costs. Each condition can be associated with specific morphological or physiologic signatures that can be identified through peripheral blood smear (PBS) analysis and other laboratory testing, allowing the machine learning (ML) architecture to detect subtle changes predictive of mortality risk. For example, conditions like DIC and acidosis can include red blood cell (RBC) shape aberrations such as schistocytes or echinocytes. The RBC encoder can detect and process the shape aberrations within the MLAtty. Dkt. No.: 115872-3344 architecture. As shown in the table, the systems and methods described herein can allow for early detection, missed diagnosis prevention, and signature-based identification could significantly improve clinical decision-making, reduce mortality, and lower costs by leveraging the automated, high-accuracy classification methods as described herein.”

[0078] FIG. 11A-C depicts example of cells that are afflicted with a condition that corresponds to a mortality of the subject. Each of the depicts are representative examples of red blood cell (RBC) shapes and white blood cell (WBC) types linked to different risk profiles. FIG. 11 A show numerous schistocytes (fragmented RBCs) and erythroblasts, alongside a teardrop cell, patterns commonly associated with microangiopathic hemolytic anemia (MAHA) and disseminated intravascular coagulation (DIC). FIG. 1 IB depicts RBCs with segmented neutrophils, band neutrophils, and macrocytes, representing comparatively benign or baseline morphologic states. FIG. 11C depicts echinocytes (spiculated RBCs) frequently seen with erythroblasts, lymphocytes, and monocytes, from samples taken from patients with echinocytosis-erythroblastosis catastrophic syndrome (EECS), acidosis, uremia, kidney failure, or liver failure.[0079| FIG. 12 depicts a graph indicating the performance of an ML model executing per condition in accordance with peripheral blood screening for early detection. FIG. 12 can include results measured by the area under the receiver operating characteristic curve (AUROC) for several hematologic diseases, such as CLL, T-LGL, MDS, MCL, and PCM. High AUROC values, such as 0.98 for CLL and T-LGL, can indicate strong discriminative capability of the ML model to identify these conditions from PBS images and possibly non-image datasets. Lower AUROC values, such as 0.80 for PCM, can reflect more challenging detection tasks where morphologic signatures may be subtler or more variable. The graph demonstrates that while the ML model consistently performs well across multiple diseases, its accuracy varies, with particularly strong results for CLL and T-LGL detection.

[0080] FIG. 13A shows a graph of relative risk of death over a seven-day period for four Al-predicted risk categories, Low, Medium, High, and Very High, with separate lines forAtty. Dkt. No.: 115872-3344 inpatient and intensive care unit (ICU) cohorts. The “Very High” risk group starts with a sharply elevated risk, peaking at around 50 times baseline on day 1, then gradually declining but remaining significantly above other groups throughout the week. This indicates that the Al model can identify patients at extreme short-term mortality risk, far exceeding average inpatient or ICU baselines.[0081 j FIG. 13B shows a graph of probability of survival over seven days for five risk categories, None, Low, Medium, High, and Very High, with separate lines for inpatient and intensive care unit (ICU) cohorts. Here, higher Al risk groups show progressively steeper drops in survival probability, with the “Very High” group falling below 70% survival within a week, while “None” and “Low” risk categories remain above 95%. The General ICU survival-to- discharge rates are 70-90%. The medical ICU survival to discharge rates are approximately 50%.[00821 FIG. 14 depicts a table presenting model performance metrics across different patient risk groups in both test and control trial validation setting. There may be separate models depending on white blood cell count (WBC) count. The models may be trained on samples in preceding 24 hours before a positive blood culture. It is shown that the model can detect bacteremia (early sepsis) for regardless of WBC count, demonstrating that the model is very useful for monitoring patients who are immuno-compromised (who cannot create increased WBCs)

[0083] FIG. 15 depicts a graph of log odd ratios for specific red blood cell (RBC) or white blood cell (WBC) morphologies. The graph compares the top 10% of morphological findings in deceased patients versus the top 10% in control patients, revealing which cellular morphology types are most strongly associated with poor outcomes. Morphologies are divided into two groups: RBC features on the left side of the plot and WBC features on the right. Among RBC morphologies, erythroblasts, schistocytes, and echinocytes exhibit the highest positive log odds ratios, indicating a strong association with mortality risk — consistent with known links to syndromes such as DIC, MAHA, and echinocytosis-erythroblastosis catastrophic syndromeAtty. Dkt. No.: 115872-3344(EECS). Conversely, normal RBC morphology and spherocytes show negative associations, implying lower mortality risk. For WBC morphologies, erythroblasts (WBC-associated identification here), blasts, metamyelocytes, and giant thrombocyte aggregates have the strongest positive log odds ratios, indicating high correlation with fatal outcomes. Cells such as lymphocytes and eosinophils show negative or near-zero odds ratios, suggesting minimal association with death.[0084J FIG. 16 depicts a graph of log odds ratios comparing the top 10% of MDS- positive cases against the top 10% of normal (control) cases, with red markers representing RBC morphologies and blue markers representing WBC morphologies. As shown, several RBC morphologies (notably echinocytes, edge cells, and spherocytes) show strong negative log odds ratios, indicating that these morphologies are significantly less common in MDS cases compared to normal controls. Among WBC morphologies, those near zero odds ratio (e.g., segmented neutrophils, eosinophils) contribute little discriminatory value for MDS detection, whereas morphologies such as blasts, erythroblasts, and unidentified atypical cells on the right side present positive log odds ratios, signifying a strong association with MDS.[0085| FIGs. 17A-C depict graphs of clustering of patient subgroups based on peripheral blood smear morphology, specifically focusing on echinocyte prevalence and its biochemical correlates. FIG. 17A is a two-dimensional scatter plot of patients projected into a reduced feature space (Component 1 vs. Component 2), with points colored by Al-identified cluster assignments. Four distinct morphological clusters are visible, plus a control group, indicating separable patient populations based on smear-derived cell features. FIG. 17B is a violin plot showing echinocyte counts (e.g., using CellaVision) per cluster. Cluster 2 stands out with markedly elevated echinocyte counts (median around 400, with values exceeding 1,200), while other clusters and controls have substantially lower counts. This confirms that the morphologic clustering correlates with quantitatively different RBC shape distributions, particularly the pathological over-representation of echinocytes. FIG. 17C is a violin plot of anion gap measurements across the same clusters. Four clusters show median anion gaps above the normalAtty. Dkt. No.: 115872-3344 reference range (shaded area), with Cluster 3 exhibiting wider variability and extreme elevations, suggesting metabolic derangements such as acidosis in association with echinocytosis.

[0086] FIG. 18A depicts a set of box plots comparing a range of morphological and laboratory features between deceased patients and controls. Each plot shows the distribution of a specific feature in both groups, along with the percentage of deceased patients exhibiting abnormal values. The multiple cell types and laboratory metrics show distributions for deceased and control groups, with percentages indicating the proportion of deceased subjects exhibiting abnormal values for each condition. Several RBC morphologies, such as echinocytes (21.2% abnormal), erythroblasts (36.0%), schistocytes (8.6%), anisocytosis (14.1%), macrocytes (14.5%), and metamyelocytes (13.4%), have higher counts or prevalence among deceased patients than controls, consistent with identification of these patterns as high-risk morphological biomarkers. Abnormal echinocyte and erythroblast counts in particular show large separations between groups, supporting their association with severe deterioration syndromes like EECS, DIC, and MAHA WBC-related features, including band neutrophils (10.9%), lymphocytes (11.4%), and artefactual cell counts (6.7-6.9%), also differ between deceased and controls, reflecting immune dysregulation or hematologic pathology detectable by smear morphology analysis. Biochemical markers such as anion gap (42.7% abnormal), albumin (22.7%), alkaline phosphatase (18.6%), alanine aminotransferase (ALT) (11.8%), aspartate aminotransferase (AST) (6%), and hematocrit (HCT) (17.2%) show significant differences between the groups.[0087J FIG. 18B depicts another set of box plots comparing a range of morphological and laboratory features between deceased patients and controls. Each plot shows distributions of a specific feature, along with the percentage of deceased patients demonstrating abnormal values. The features span hematology, morphology, and machine learning outputs. Blood urea nitrogen (BUN) is markedly elevated in deceased patients (47.3% abnormal), reflecting renal dysfunction and metabolic derangement. Platelet counts (PLT) are significantly lower in deceased patients (18.8% abnormal), consistent with consumptive coagulopathy syndromes such as DIC. Red blood cell count (RBC) and polychromatic RBCs show smaller but notable differences between groups, suggesting altered erythropoiesis in at-risk patients. Morphologic cell features includeAtty. Dkt. No.: 115872-3344 absolute monocyte counts (higher in deceased group), myelocytes (13.3% abnormal), and anisocytosis measured by red cell distribution width (RDW) (16.8% abnormal), all pointing to reactive or dysplastic hematologic changes. The MIL score shows extremely high values for deceased patients (75.1% abnormal) compared to controls.

[0088] FIG. 19 depicts a graph of ranking of predictive importance of individual features for mortality patients % outside 95% confidence interval versus control. The features encompass blood smear morphology and chemistry panel measurements. As shown, biochemical values such as BUN (blood urea nitrogen), anion gap, and CO2 levels, all indicators of metabolic disturbances and organ dysfunction. High-ranking morphology features include echinocyte count and erythroblast percentage, reflecting abnormal RBC shapes and immature red cell presence strongly linked to acute deterioration syndromes. Other prominent contributors include bilirubin, AST (aspartate aminotransferase), potassium, and differential counts for specific WBC subtypes like segmented neutrophils. Additional morphology types (e.g., schistocytes, macrocytes, myelocytes) and lab parameters (e.g., hematocrit, RDW) provide incremental predictive value. Features near the bottom of the chart have low individual importance, including certain rare morphologies (e.g., spherocytes, giant thrombocyte aggregates, unidentified cells) that contribute minimally in isolation.[0089J FIG. 20 depicts a graph of ranking of predictive importance of individual features based on the Interquartile Range (IQR) Outlier method. As shown, morphological counts of erythroblasts and blasts, which are strong indicators of severe hematologic and systemic pathology. Elevated anion gap and CO2 abnormalities also rank highly, reflecting metabolic derangements such as acidosis commonly present in critical illness. Other major contributors include metamyelocyte counts, creatinine, and hypochromatic RBCs, which can indicate renal impairment, bone marrow stress, and red cell production abnormalities. Mid-tier predictors include additional morphologic features such as macrocytes, schistocytes, immature myeloid forms (myelocytes, promyelocytes), and biochemical values (bilirubin, AST), all of which can signal underlying syndromes like sepsis, DIC, MAHA, or EECS. Lower-ranked features toward the right side of the chart, including rare morphologies (e.g., spherocytes, giantAtty. Dkt. No.: 115872-3344 thrombocyte aggregates, plasma cells) and certain standard labs (e.g., hemoglobin, hematocrit), contribute less individually.B. Systems and Methods for Determining Values Indicating Probabilities of Mortality of Subjects Using Blood Samples

[0090] Evaluating morphology of blood cells can involve modalities, such as peripheral pap smear, complete blood count, flow cytometry, bone marrow aspiration, automated digital morphology, and reticulocyte count which vary in terms of speed, automations, and convenience, among others, to assess the quality and quantity of the blood cells while avoiding manual microscopic examination. From these data, it is possible to obtain information associated with the health and function of the blood cells. The information can be used in a variety of applications, such as diagnosis, treatment selection, mortality evaluation, and labeling datasets to train artificial intelligence (Al) architectures to determine values indicating probabilities of mortality of subjects.

[0091] Methods to detect deterioration of survival of subjects may rely on subjective assessments of these pieces of data. Under one approach, a human (e.g., lab technician) may examine images of each individual blood smear under a microscope to determine the morphology of the blood cells. Such manual techniques may be tedious and time-consuming and may rely on specialized expertise and knowledge, thereby limiting their accessibility. These approaches also may miss subtle indicators present in blood morphology, and may result in serious acute syndromes (e.g., sepsis, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MAHA), and acute metabolic derangement) going undetected until the subject is critically ill. Automated blood count systems cannot detect these subtle morphological changes, and manual slide reviews are time-intensive and inconsistently performed. This results in delayed intervention and preventable deaths.10092] To address these and other technical challenges, a machine learning technique can be used to determine values indicating probabilities of conditions (e.g., deterioration orAtty. Dkt. No.: 115872-3344 mortality) for subjects. The machine learning architecture may employ multimodal, multiencoder, and multiple instance learning (MIL) framework to aggregate image and non-image data to determine values for conditions of subjects. A computing system can receive an images of red blood cells (RBCs) and white blood cells (WBCs) as well as non-image data (e.g., complete blood count (CBC), chemistry panel, physiological measurements, and other clinical data) for a given subject. The computing system can feed the RBC and WBC images along with the non-image data to the ML architecture. From feeding, the ML architecture can process the RBC and WBC images and non-image data to generate embeddings. The embeddings can identify latent features (e.g., abnormal morphologies) correlated with deterioration and mortality. The ML architecture can also process the embeddings to output a value to indicate a probability of a condition of the subject. The outputs may be provided in an interpretable formal directly to a computing device to direct further care of the subject, such as treatment administration, triage, escalation of care, protocols, or discharge planning, among others.

[0093] In this manner, the ML architecture can combine image data with non-image data to generate earlier and more accurate detection of deterioration or mortality in the subject. From a clinical perspective, by leveraging multimodal data, the ML architecture can identify acute syndromes before they would be apparent through vital signs or other physiological measurements. This earlier alerting allows for timely escalation of care, admission when warranted, and rapid initiation of targeted diagnostic testing and therapeutic interventions. As a result, the outputs of the ML architecture may be used to reduce mortality rates, lower the incidence of unexpected ICU transfers, and prevent multi-organ failure by addressing primary causes before downstream complications arise. From a computer perspective, the integration of multimodal data can enable extraction of latent features correlated with deterioration or morality. In addition, the provision of more accurate and earlier values indicating probabilities of conditions can yield a more efficient use of computing resources (in terms of processor and memory), relative to approaches in which computing resources are wasted from providing less accurate outputs.Atty. Dkt. No.: 115872-3344[0094| Referring now to FIG. 21, depicted is a block diagram of a system 100 of determining values indicating probabilities of mortality of subjects using blood samples. In a brief overview, the system 100 can include at least one data processing system 105, at least one image capture device 110, at least one administrative device 115, and at least one database 155, among others, communicatively coupled via at least one network 120. The data processing system 105 can include at least one dataset indexer 125, at least one image sorter 130, at least one model trainer 135, at least one model applier 140, at least one output evaluator 145, and at least one machine learning (ML) architecture 150, among others. Each of the components of the system 100 can be implemented using the computing system as described in Section C. The system 100 may be used to implement the functionalities as described in Section A.[0095| In further detail, the data processing system 105 can be any computing device comprising one or more processors coupled with memory and software capable of performing the various processes and tasks described herein. The data processing system 105 can be housed within a computing system (e.g., laptop, PC, smart device) or within a server group (e.g., a data center, a branch office, or a server site), and include instructions to manage the identifying of images, generating a classification, and storing an association. The data processing system 105 can be in communication with the imaging device 110, administrative device 115, and the database 155, among others.

[0096] The data processing system 105 can include or execute any number of modules, processes, components, or subcomponents to perform the various processes and tasks described herein. On the data processing system 105, the dataset indexer 125 within the data processing system 105 can receive, retrieve, or otherwise identify a dataset from the imaging device 110 and the administrative device 115. The image sorter 130 can receive, select, or otherwise identify white blood cell (WBC) images and red blood cell (RBC) images. The model trainer 135 can train, establish or otherwise initialize the ML architecture 150. The model applier 140 can apply, execute, or otherwise the WBC images and the RBC images to the feed the ML architecture. The output evaluator 145 can generate, determine, or otherwise identify an output based on a classification.Atty. Dkt. No.: 115872-3344[0097| The ML architecture 150 can be any type of ML algorithm or model to determine a value indicating a probability of condition (e.g., mortality or deterioration) for a subject based on the red blood cells (RBCs) and the white blood cells (WBCs). The ML architecture 150 can be maintained on the data processing system 105. The architecture can be, for example, a deep learning artificial neural network (ANN) such as an encoder-decoder model with a convolution neural network architecture, a transformer architecture, a diffusion model, or an encoder-classifier model (e.g., as depicted in FIG. 1), among others. In general, the ML architecture 150 can have an image of RBC or WBC in any modality from a subject as an input and a determination of a value indicating a probability of condition (e.g., mortality or deterioration) as an output. The ML architecture 150 may have been initialized, trained, and established using training data in accordance with learning techniques (e.g., supervised or semi-supervised). The training data can include or identify images or non-image data set and a label for the subject. In some embodiments, the ML architecture 150 can be trained for a specific subject. In some embodiments, the ML architecture 150 can be trained for a group of subj ects and then refined or fine-tuned for a particular subject.[0098J The ML architecture 150 can include at least one WBC encoder 160, at least one RBC encoder 165, at least one feature encoder 170 at least one aggregate predictor 175 (“agg predictor 175” as depicted), at least one classifier 180, and at least one clustering model 190, among others. The WBC encoder 160 can execute dimensionality reduction, feature extraction, or sequence modeling to generate, determine, or otherwise create a set of WBC embeddings. The RBC encoder 165 can execute dimensionality reduction, feature extraction, or sequence modeling to generate, determine, or otherwise create a set of RBC embeddings. The feature encoder 170 can execute dimensionality reduction, feature extraction, or sequence modeling to generate, determine, or otherwise create a set of feature embeddings The aggregate predictor 175 can generate, determine, or otherwise identify a value indicating a probability of condition for the subject. The classifier 180 can generate, determine, or otherwise identify a classification for the value indicating the probability of condition.Atty. Dkt. No.: 115872-3344[0099| The clustering model 190 can generate, determine, or otherwise identify a plurality of assignments based on the RBC embeddings, the WBC embeddings, and the feature embeddings. The clustering model 190 can include k-means clustering, a density -based spatial clustering of applications with noise (DBSCAN), hierarchical clustering, or Gaussian mixture model (GMM), among others. The clusters can be pre-defined using labeled data or be generated using unlabeled training data. The feature space can be ^-dimensional corresponding to the number dimensions of the embedding sets outputted by the feature encoder 170. The assignment of the embedding sets to clusters can be used for various inferences or statistical analyses.[01001 The imaging device 110 can be any device capable of acquiring images of blood samples of subjects. The image can be captured via various techniques such as fluorescence microscopy, phase-contrast microscopy, bright-field microscopy, confocal microscopy, scanning electron microscopy, transmission electron microscopy, quantitative phase imaging, and automated digital microscopy, among others. Although primarily discussed herein in terms of peripheral blood smear (PBS), other imaging modalities besides those listed above may be supported by the data processing system 105. The imaging device 110 can be in communication with the data processing system 105 and the administrative device 115 to provide acquired images.[0101 | The administrative device 115 can be any device comprising one or more processors coupled with memory and software and capable of providing an output projection image. The administrative device 115 can be associated with an entity (e.g., clinician, physician, doctor) examining the subject or biomedical images from the subject. The administrative device 115 can be in communication with the data processing system 105 and the imaging device 110 to exchange data. The administrative device 115 can display images acquired from the imaging device 110 on a display.

[0102] Referring now to FIG. 22A, depicted is a block diagram of a process 200 of training a machine learning architecture (ML) to determine a presence or absence of a condition indicating an expected mortality of subjects using blood samples. The process 200 can include or correspond to operations performed in the system 100. Under the process 200, the dataset indexerAtty. Dkt. No.: 115872-3344125 can receive, retrieve, or otherwise extract training data 205 from the database 155. The training data 205 can be for a subject 210 and an associated blood sample 215 of the subject 210. The training data 205 can include a plurality of examples. Each example of the training data 205 can include sample images 220A-N (sometimes referred to as images 220), a non-image dataset 225 associated with the subject 210, and at least one label 230 for the subject 210.

[0103] The subject 210 (sometimes herein referred to as a sample subject 210 when associated with the training data 205) can be a human or animal subject, among others. The subject 210 can have, can be at risk of, can be afflicted, can be diagnosed with cancer. The cancer can be at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases, among others.

[0104] The training data 205 for the subject 210 can include the images 220 can derived from at least one blood sample 215 from the subject 210. The images 220 can captured from the imaging device 110 and stored within the database 155. A physician can extract, obtain, or otherwise retrieve a blood sample 215 from the subject 210. The blood sample 215 may be acquired in accordance with a peripheral blood smear (PBS). The blood sample 215 can be extracted from peripheral veins, a central venous catheter, or bone marrow of the subject 210 for analysis. Once extracted, the physician can use the imaging device 110 to capture the images 220 of the blood sample 215. The blood sample 215 can include various compositions of the blood, such as, circulating tumor cells, leukemic cells, prostate-specific antigens, CA-125, Carcinoembryonic Antigen, Alpha-fetoprotein, CA 19-9, altered blood cell counts (e.g., low redAtty. Dkt. No.: 115872-3344 blood cells, low white blood cells) C-reactive Proteins, among other compositional changes (e.g., due to cancer or other medical conditions). The blood sample 215 can include RBC, WBC, plasma, platelets, enzymes, minerals, glucose, lipids, serum, blood proteins, among others. The imaging device 110 can store the images 220 within the database 155 for use as training data 205. The images 220 may be acquired in accordance with any number of imaging techniques, such as fluorescence microscopy, phase-contrast microscopy, bright-field microscopy, confocal microscopy, scanning electron microscopy, transmission electron microscopy, quantitative phase imaging, and automated digital microscopy, among others.

[0105] Within the training data 205, the non-image dataset 225 can include characteristics, information, and traits about the subject 210. The traits of the subject 210 can be provided by the subject 210, extracted from an electronic medical record, or documented by a physician. The traits can include symptoms (e.g., fatigue, weakness, bleedings), medical history, family medical history, lab results, physical traits, psychological traits (e.g., personality, coping mechanism, emotional state), behavioral traits (e.g., adherence to treatment, health seeking behavior, communication style), social traits, among other traits. The non-image dataset 225 can include a blood count derived from the blood sample 215 of the subject 210. The blood count can be a complete blood count, RBC count, WBC count, hemoglobin, platelet count, among others. In some embodiments, the blood count may indicate an estimated total blood count in the body of the subject 210. For example, the blood count may be derived as a function of the blood count in the blood sample 215. The non-image dataset 225 can further include a plurality of parameters derived from testing of the blood sample 215. The plurality of parameters (e.g., chemistry panel parameters) can include one or more of: glucose level, calcium level, sodium level, potassium level, chloride level, carbon dioxide level, blood urea nitrogen (BUN), a creatinine level, albumin level, total protein level, alkaline phosphatase (ALP) level, alanine aminotransferase (ALT) level, aspartate aminotransferase (AST) level, anion gap, or bilirubin level, among others. The non- image dataset 225 can include a plurality of physiological measurements of the subject 210. The plurality of physiological measurements can include vital signs, temperature, heart rate, respiratory rate, blood pressure, oxygen saturation, age, height, weight, body mass index (BMI), and bodyAtty. Dkt. No.: 115872-3344 surface area (BSA), among others. Within the training data 205, the label 230 can correspond to the classification of the subject 210. For example, the label 230 can indicate that the subject 210 is at risk of dying within a time interval relative to the time of the extraction of the blood sample 215. The time interval can be any range between 6 hours to 6 months, for example, at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 60days, 90 days, 120 days, or 180 days, among other time intervals. In another example, the label 230 can indicate that the subject 210 is to survive for the time interval relative to the time of the extraction of the blood sample 215. In some embodiments, the label 230 may identify a presence or an absence of at least one of a plurality of conditions affecting the subject 210. The conditions may include, for example, a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MAHA), acidosis, multi-organ failure, anemia, hemophagocytic lymphhistiocytosis, or cytokine release syndrome, among other conditions associated with the mortality of the subject 210. For each condition, the label 230 may, for example, an indication of a presence or absence of the condition.[0106J In some embodiments, the label 230 can identify or indicate a clinical decision to be performed with respect to the subject 210. The clinical decision may include, for example, any one or more of the following: (1) admission level (e.g., deciding between general admission, ICU, or step-down unit based on stability); (2) monitoring frequency (e.g., determining the need for continuous vs. intermittent vital signs monitoring); (3) diagnostic testing (e.g., choosing between non-invasive tests (like ultrasound) and more invasive or strenuous procedures such as CT scans with contrast); (4) medication adjustment (e.g., adjusting dosages or types of medications, especially for sedatives, anticoagulants, or medications affecting heart rate and blood pressure); (5) surgical timing (e.g., postponing or expediting surgery based on a patient’s hemodynamic stability); (6) transport decisions (e.g., deciding if a patient is stable enough to be moved within the hospital such as for imaging or a specialist visit); (7) nutritional support (e.g., deciding whether to feed orally, via enteral tube, or parenterally); (8) ventilation support (e.g., determining if the patient needs additional respiratory support, such as supplemental oxygen or mechanicalAtty. Dkt. No.: 115872-3344 ventilation); (9) discharge planning (e.g., assessing whether a patient can be discharged, needs a referral to a skilled nursing facility, or requires home health support); (10) fluid management (e.g., tailoring IV fluid administration to avoid overloading or under-hydrating the patient); (11) blood transfusions (e.g., deciding when or if to transfuse blood products, especially if there’s instability in hematocrit or hemoglobin levels); (12) physical activity and rehabilitation (e.g., determining the level and timing of physical therapy or mobility interventions); (13) consultations (e.g., prioritizing specialist consultations, such as cardiology or nephrology, based on stability and risk); (14) code status discussion (e.g., engaging in end-of-life discussions and potentially revising code status based on a decline in stability); (15) sepsis and infection management (e.g., prompt initiation or escalation of antibiotics in unstable patients, given their higher susceptibility); (16) pain management (e.g., tailoring the type and dose of analgesics to avoid respiratory depression or hypotension in unstable patients); (17) fluid restriction (e.g., implementing or adjusting fluid restrictions, particularly in cases of heart or kidney failure where instability can worsen with fluid overload); (18) medication route (e.g., deciding whether medications should be administered orally, intravenously, or subcutaneously based on the patient’s stability and ability to tolerate each method); (19) sedation and anesthesia level (e.g., choosing lighter sedation for procedures in unstable patients who might not tolerate deep sedation or general anesthesia); (20) anti coagulation management (e.g., modifying anticoagulation therapy, balancing the risk of bleeding with the risk of clot formation, particularly if the patient is hemodynamically unstable); (21) vital signs review frequency (e.g., increasing the frequency of nursing assessments and vital sign checks in unstable patients to detect any rapid changes); (22) electrolyte management (e.g., prompt correction of abnormal electrolytes such as potassium or magnesium to stabilize cardiac and muscular function in fragile patients); (23) infection control precautions (e.g., placing unstable patients in isolation more promptly to prevent infections that could further destabilize their condition); (24) blood pressure management protocol (e.g., adjusting medications to prevent drops in blood pressure, especially in patients with low perfusion risk); (25) family visitation (e.g., deciding on restricted or flexible visitation hours to avoid overstimulation or to allow family support in unstable situations, especially in ICU settings); (26) ambulation restrictions (e.g., limiting or delaying ambulation to prevent falls or exertion-related complications in unstable patients); (27)Atty. Dkt. No.: 115872-3344 temperature management (e.g., implementing cooling or warming protocols for patients who are hemodynamically unstable or experiencing extreme fever or hypothermia); (28) electrocardiogram (ECG) monitoring (e.g., deciding whether continuous ECG monitoring is necessary to track cardiac rhythm in patients with unstable vital signs); (29) risk of pressure injuries (e.g., implementing repositioning schedules, special mattresses, or wound care in patients with poor mobility or circulation to prevent pressure sores); (30) end-of-life or palliative care initiatives (e.g., re-evaluating goals of care and possibly initiating palliative care when patient stability declines significantly and recovery seems unlikely); (31) ICU bed allocation (e g., prioritizing ICU or high- dependency unit beds for patients whose stability is at higher risk of deterioration); or (32) emergency dialysis initiation (e.g., deciding if immediate dialysis is required for unstable patients with electrolyte imbalances or fluid overload), among others.

[0107] In some embodiments, with multiple subjects 210, the dataset indexer 125 can index, sort, or otherwise organize each example within then training data 205 to correspond to each subject 210 in the plurality of subjects 210. The dataset indexer 125 may process or parse each example in the training data 205 to identify or extract the images 220, the non-image dataset 225, and the label 230, among others, for the corresponding sample subject 210. In some embodiments, the dataset indexer 125 may partition, section, or otherwise divide a single image 220 into a set of patches forming the images 220. Each patch may correspond to a respective portion of the image 220. For example, the training data 205 may originally include a single image 220, from which the dataset indexer 125 divides to form the plurality of images 220 for analysis.(0108| The image sorter 130 can identify, determine, or otherwise select WBC images 220'A-N (sometimes referred to as WBC images 220') from the images 220. The selection may be based on a visual characteristic of the image 220. The image sorter 130 can analyze each image 220 to identify the WBC images 220'. In operation, the image sorter 130 can obtain each image 220 from the plurality of images 220 from the dataset indexer 125. Upon reception of an image 220, the image sorter 130 can preprocess the image. While preprocessing, the image sorter 130 can remove noise, distortions, artifacts, or stains associated with the image 220 and perform contrast enhancement (e.g., adaptive contrast stretching, histogram equalization) between the cellsAtty. Dkt. No.: 115872-3344 of the blood sample 215 and the background of the image 220. Once each image 220 is preprocessed, the image sorter 130 can execute segmentation to identify at least one portion, region, or object of the image 220 to separate the WBCs from the background and non-similar cells (e.g., a WBC from a group of RBCs). The image sorter 130 can use thresholding, algorithms (e.g., watershed algorithm), or edge detection, among other methods to execute segmentation.

[0109] Continuing on, the image sorter 130 can extract features (e.g., visual characteristics) from each cell in the image 220 to distinguish the WBCs from the RBCs. The features can include, but are not limited to, shape, size, color, intensity, nucleus to cytoplasm ratio, granularity, among other distinguishing features. In some embodiments, the image sorter 130 can use a ML algorithm to classify the WBCs. The ML algorithm can be, for example, support vector machine (SVM), Random Forest, K-Nearest Neighbor (KNN), deep learning CNN, another other algorithms. Once classified, the image sorter 130 can identify the WBC image 220' as an image 220 that mostly (e.g., at least 50-70%) contains WBCs from the images 220 of the blood sample 215.

[0110] Concurrently, the image sorter 130 can identify, determine, or otherwise select RBC images 220"A-N (sometimes referred to as RBC images 220"). The selection may be based on a visual characteristic of the image 220. The image sorter 130 can analyze each image 220 to identify the RBC images 220". In operation, the image sorter 130 can obtain each image 220 from the plurality of images 220 from the dataset indexer 125. Upon reception of an image 220, the image sorter 130 can preprocess the image. While preprocessing, the image sorter 130 can remove noise, distortions, artifacts, or stains associated with the image 220 and perform contrast enhancement (e.g., adaptive contrast stretching, histogram equalization) between the cells of the blood sample 215 and the background of the image 220. Once each image 220 is preprocessed, the image sorter 130 can execute segmentation to identify at least one portion, region, or object of the image 220 to separate the RBC images 220" from the background and non-similar cells (e.g., an RBC from a group of WBCs). The image sorter 130 can use thresholding, algorithms (e.g., watershed algorithm), or edge detection, among other methods to execute segmentation.Atty. Dkt. No.: 115872-3344[01111 Continuing on, the image sorter 130 can extract features (e.g., the visual characteristics) from each cell in the image 220 to distinguish the RBCs from the WBCs. The features can include, but are not limited to, shape, size, color, intensity, nucleus to cytoplasm ratio, granularity, among other distinguishing features. In some embodiments, the image sorter 130 can use a ML algorithm to classify the RBCs. For example, depending on the staining of the sample, the RBC images 220" may have a redder hue than WBC images 220’ . The ML algorithm can be, for example, support vector machine (SVM), Random Forest, K-Nearest Neighbor (KNN), deep learning CNN, another other algorithms. Once classified, the image sorter 130 can identify the RBC image 220" as an image 220 that mostly contains RBCs (e g., at least 50-70%) from the images 220 of the blood sample 215.[01121 The model trainer 135 can apply, feed, or otherwise input the WBC images 220' to the ML architecture 150 for each example of the training data 205. The model trainer can apply, execute, or otherwise use the ML architecture 150 on the training data 205 for the one or more subjects 210. The ML architecture 150 can include the at least one WBC encoder 160, the at least one aggregate predictor 175 and at least one classifier 180, among others. The WBC encoder 160 can ingest, received or otherwise obtain the WBC images 220' from the model trainer 135. The ML architecture 150 can include a plurality of parameters for determine the value indicating the probability of condition. In some embodiments, the plurality of parameters can include a plurality of hyperparameters to establish the architecture, design, or configuration of the ML architecture 150 to determine the value indicating the probability of condition.[0U3| Within the ML architecture 150, the plurality of hyperparameters can be arranged across the WBC encoder 160, the RBC encoder 165, the aggregate predictor 175, and the classifier 180. The plurality of hyperparameters can include a learning rate, number of epochs, a batch size, a model architecture (e.g., deep learning convolutional neural network, ensemble network, or clustering algorithm, or any combination thereof), and regularization parameters. In some implementations, the plurality of parameters can include a plurality of model parameters. The plurality of model parameters can be variables learned from the training data 205 (i.e., one or more examples) that establish, dictate, or otherwise define a link between the input (e.g., WBC imagesAtty. Dkt. No.: 115872-3344220’) and the output (e.g., classification 245). The model parameters can continuously optimize during training to minimize a loss function of the ML architecture 150. The plural of model parameters can include target weights or biases (e.g., influence the output of the ML architecture 150 based on accurate targets), loss function (e.g., compound loss), among others. This is not limited to the WBC encoder 160, but the hyperparameters of the ML architecture 150 can be arranged and adjusted for the RBC encoder 165 in a similar manner.[0114 In feeding, the WBC encoder 160 can generate, produce, or otherwise determine a set of WBC embeddings 235A-N (sometime referred to as embeddings 235). To generate the set of WBC embeddings 235, the WBC encoder 160 can execute dimensionality reduction, feature extraction, or sequence modeling. In some implementations, the WBC encoder 160 can include a plurality of layers to extract, retrieve, or otherwise obtain the WBC images 220' from the image sorter 130 and the model trainer 135. The plurality of layers can include one or more hidden layers. The one or more hidden layers can transform the WBC images 220'to extract or retrieve one or more representations of the WBC images 220'. The one or more representations can include the set of WBC embeddings 235.[0U5| Concurrently, the model trainer 135 can apply, feed, or otherwise input the RBC images 220" to the ML architecture 150 for each example of the training data 205. The model trainer can apply, execute, or otherwise use the ML architecture 150 on the training data 205 for the one or more subjects 210. The ML architecture 150 can include the at least one RBC encoder 165, the at least one aggregate predictor 175 and at least one classifier 180, among others. The RBC encoder 165 can ingest, receive, or otherwise obtain the RBC images 220" from the model trainer 135. The ML architecture 150 can include a plurality of parameters for determine the value indicating the probability of condition. In some embodiments, the plurality of parameters can include a plurality of hyperparameters to establish the architecture, design, or configuration of the ML architecture 150 to determine the value indicating the probability of condition.[0U6J Within the ML architecture 150, the plurality of hyperparameters can be arranged across the RBC encoder 165, the RBC encoder 165, the aggregate predictor 175, and the classifierAtty. Dkt. No.: 115872-3344180. The plurality of hyperparameters can include a learning rate, number of epochs, a batch size, a model architecture (e.g., deep learning convolutional neural network), and regularization parameters. In some implementations, the plurality of parameters can include a plurality of model parameters. The plurality of model parameters can be variables learned from the training data 205 (i.e., one or more examples) that establish, dictate, or otherwise define a link between the input (e.g., RDC images 220") and the output (e.g., classification 245). The model parameters can continuously optimize during training to minimize a loss function of the ML architecture 150. The plurality of model parameters can include target weights or biases (e.g., influence the output of the ML architecture 150 based on accurate targets), loss function (e.g., compound loss), among others. This is not limited to the RBC encoder 165, but the hyperparameters of the ML architecture 150 can be arranged and adjusted for the WBC encoder 160 in a similar manner. From feeding, the RBC encoder 165 can generate, produce, or otherwise determine a set of RBC embeddings 235’ A-N (sometime referred to as embeddings 235’). To generate the set of RBC embeddings 235’, the RBC encoder 165 can execute dimensionality reduction, feature extraction, or sequence modeling. In some implementations, the RBC encoder 165 can include a plurality of layers to extract, retrieve, or otherwise obtain the RBC images 220" from the image sorter 130 and the model trainer 135. The plurality of layers can include one or more hidden layers. The one or more hidden layers can transform the RBC images 220"to extract or retrieve one or more representations of the RBC images 220". The one or more representations can include the set of RBC embeddings 235’.

[0117] The model trainer 135 can apply, feed, or otherwise input the non-image dataset 225 to the feature encoder 170 for each example of the training data 205. The model trainer 135 can apply, execute, or otherwise use the ML architecture 150 on the non-image dataset 225 for the one or more subjects 210. The ML architecture 150 can include the at least one feature encoder 170, the at least one aggregate predictor 175 and at least one classifier 180, among others. The feature encoder 170 can ingest, receive, or otherwise obtain the non-image dataset 225 from the model trainer 135. From feeding, the feature encoder 170 can generate, establish or otherwise determine a set of feature embeddings 235” A-N (referred to as embedding 235” herein). TheAtty. Dkt. No.: 115872-3344 feature encoder 170 (similar to the RBC encoder 165 and the WBC encoder 160) can execute one or more of dimensionality reduction, feature extraction, on the data within the non-image dataset 225. The model trainer 135 can transform the non-image dataset 225 into a feature vector prior to ingestion by the feature encoder 170. In this manner, the feature encoder 170 can execute the functions (e.g., feature extraction) on the feature vector. The feature encoder 170 can include a plurality of layers to extract, retrieve, or otherwise obtain the non-image dataset 225 from the model trainer 135 and the image sorter 130. The plurality of layers can include an input layer, hidden layers (e.g., a fully connected layer, batch normalization layer, activation layer), and an output layer. The input layer can receive the feature vector or transform the non-image dataset 225 into the feature vector. The hidden layers can process the feature vector by applying a linear transformation and execute a non-linear function on each aspect of the non-image dataset 225 (e.g., traits, parameters) to generate intermediary output values. Each of the intermediary output values can indicate patterns or connection for the output layers to establish the final embeddings 235. The output layer can generate each of the embeddings 235” using the intermediary output values.[0118| Upon generation of the embeddings 235, the WBC encoder 160 can provide, feed, or otherwise input the embeddings 235 into the aggregate predictor 175. In a similar manner, upon generation of the embeddings 235’, the RBC encoder 165 can transmit, feed, or otherwise input the embeddings 235’ to the aggregate predictor 175. In a similar manner, the feature encoder 170 can provide, feed, or otherwise input the embeddings 235” to the aggregate predictor 175. The aggregate predictor 175 can receive, ingest, or otherwise obtain the embeddings 235, the embeddings 235’, and the embeddings 235”. The aggregate predictor 175 can generate, determine, or otherwise predict a value 240 indicating a probability of mortality (or the condition) for the subject 210 based on the embeddings 235 and the embeddings 235’. The aggregate predictor 175 can generate, determine, or otherwise predict a value 240 indicating a probability of a condition of the plurality of conditions associated with the mortality for the subject 210 based on the embeddings 235, the embeddings 235’, and the embeddings 235”. The aggregate predictor 175 can execute, for example, bagging (e.g., random forests), boosting (e.g., gradient boosting), or stacking to determine the value 240 indicating the probability of condition.Atty. Dkt. No.: 115872-3344[0119| From the execution, the aggregate predictor 175 can generate the value 240 based on at least one of the embeddings 235, the embeddings 235’, and the embeddings 235”. The value 240 may indicate the probability of mortality for the subject 210 for the given time interval. The probability of mortality can indicate a likelihood that the subject 210 can die based on the condition of the RBC and WBC within the RBC images 220" and the WBC images 220' for at least the given time period. In some embodiments, with the determination of the probability for each of the RBC and WBC images, the aggregate predictor 175 can combine the predictions to determine the value 240 indicating the probability of mortality. To determine the value 240, the aggregate predictor 175 can implement averaging, majority voting, weighted averaging, meta-model in stacking, among other methods to determine the value 240.[0120| In some embodiments, the aggregate predictor 175 can generate the value 240 indicating a probability of the presence (or absence) of a condition of the plurality of conditions associated with the mortality for the subject 210 based on at least one of the embeddings 235, the embeddings 235’, and the embeddings 235”. The probability of the condition can indicate a likelihood of the presence of the condition in the subject 210. The probability of the condition can be independent of time. In some embodiments, the aggregate predictor 175 can generate a plurality of values 240 corresponding to the plurality of conditions based on at least one of the embeddings 235, the embeddings 235’, and the embeddings 235”. For each condition, the value 240 can indicate the probability of the presence (or absence) of the condition in the subject 210. In some embodiments, the aggregate predictor 175 can determine or generate the value 240 to indicate the likelihood of clinical decision to be carried out for the subject 210 based on the embeddings 235, the embeddings 235’, and the embeddings 235”. For each candidate clinical decision, the aggregate predictor 175 can determine or generate the value 240 to indicate likelihood of taking the candidate clinical decision.[0121 | The aggregate predictor 175 can feed, input, or otherwise provide the value 240 indicating condition to the classifier 180. Concurrently, the model trainer 135 can feed, input or otherwise provide the non-image dataset 225 to the classifier 180. The classifier 180 can include one or more input layers that receives the value 240 indicating the probability of mortality and theAtty. Dkt. No.: 115872-3344 non-image dataset 225 to provide the value 240 and the non-image dataset 225 to the subsequent layers of the classifier 180. In some embodiments, the classifier 180 can execute, for example, bagging (e.g., random forests), boosting (e.g., gradient boosting), or stacking to determine the value 240 indicating the probability of mortality for the time interval. In some embodiments, the classifier 180 can include one or more hidden layers (e.g., fully connected layers, convolutional layers, recurrent layers) to learn patterns and representations of the value 240 with the non-image dataset 225 based on an activation function (e.g., ReLU, Sigmoid, Tanh). For instance, the one or more hidden layers can form a pattern based on the value 240 and a trait of the subject 210 within the non-image dataset 225.

[0122] Using the patterns and representations, the output layer of the classifier 180 can provide at least one classification 245 for the subject 210. The output layer can be at least one of binary classification or multi class classification. The classification 245 for the subject 210 can be at least one of mortality or survival for the subject 210 in the given time interval based on the inputs of the value 240 and the non-image dataset 225. In some embodiments, in processing the inputs, the classifier 180 may calculate, generate, or otherwise determine an intermediary value. The value may indicate a probability or mortality for the subject 210 based on the value 240 calculated from the images 220’ and 220” and the non-image dataset 225. The classifier 180 may compare the intermediary value to a threshold. If the value satisfies (e.g., greater than or equal to) the threshold, the classifier 180 may generate the classification 245 to indicate that the subject 210 is at risk of dying for at least the given time interval. Otherwise, if the value does not satisfy (e.g., less than) the threshold, the classifier 180 may generate the classification 245 to indicate survival for at least the given time interval. When the classification 245 corresponds to mortality, the subject 210 can be at risk of dying because of the cancer and / or condition. However, when the classification 245 corresponds to survival, the subject 210 is not at risk of dying to the cancer and / or condition.[0123| In some embodiments, the classifier 180 can generate the classification 245 using the value 240 indicating the probability of the condition associated with the mortality of the subject 210. The classifier 180 can use the value 240 to generate the classification indicating an absenceAtty. Dkt. No.: 115872-3344 or a presence of the condition. The classifier 180 can generate an intermediary value using the value 240. The intermediary value can be a binary, flag, or marker that indicates the respective condition impacting the blood sample 215 of the subject 210. The classifier 180 may compare the intermediary value to a threshold. If the value 240 satisfies (e.g., greater than or equal to) threshold, the classifier 180 can generate the classification 245 to indicate the presence of the condition. On the other hand, if the value 240 does not satisfy (e.g., less than) threshold, the classifier 180 can generate the classification 245 to indicate the absence of the condition. The classifier 180 can iterate through the set of values 240 for the various conditions.

[0124] In some embodiments, the classifier 180 can generate the classification 245 to indicate which clinical decision is to be carried out for the subject 210 based on the value 240. The classifier 180 can generate an intermediary value using the value 240. The intermediary value can be a binary, flag, or marker that indicates the respective clinical decision to be taken (e.g., to effectuate the proper procedure or action to not take on the subject 210). The classifier 180 may compare the intermediary value to a threshold. If the value 240 satisfies (e.g., greater than or equal to) the threshold, the classifier 180 can generate the classification 245 to indicate the selection of the clinical decision. On the other hand, if the value 240 does not satisfy (e g., less than) the threshold, the classifier 180 can generate the classification 245 to indicate the exclusion of the clinical decision.

[0125] The model trainer 135 can initialize, train, or otherwise establish the ML architecture 150 for each subject 210 interacting with the imaging device 110 by retrieving, obtaining, or otherwise receiving the training data 205 from the dataset indexer 125. The model trainer 135 may transmit the training data 205 to the image sorter 130 to feed the ML architecture 150 the WBC images 220' and the RBC images 220" to determine the value 240 and generate the classification 245. In some embodiments, the model trainer 135 can train the ML architecture 150 for each subject 210 to enable a single ML model to execute per subject 210. In some embodiments, the model trainer 135 can initialize, train, or otherwise establish the ML architecture 150 for a set of subjects 210. The set of subjects 210 may correspond to a cohort or a population of subjects with common characteristics (e.g., similar cancer, similar traits).Atty. Dkt. No.: 115872-3344[0126| To update the parameters of the ML architecture 150, the model trainer 135 may determine, calculate, or otherwise generate at least one least one loss metric 250 based on the classification 245. In some implementations, the model trainer 135 may generate the loss metric 250 be based on the value 240. In some implementations, the classifier 180 can generate the loss metric 250. The loss metric 250 can measure, rate, or identify how well the ML architecture 150 can determine accurate classifications 245 for the subjects 210. The loss metric 250 from the classifier 180 can measure discrepancy between the classification 245 and the label 230. In further detail, when classifier 180 is presented with the value 240 indicating a high probability of mortality, the classifier 180 can learn to generate a classification 245 of mortality for the subject 210 and calculate the loss metric 250. Conversely, when classifier 180 is presented with the value 240 indicating a law probability of mortality, the classifier 180 can learn to generate a classification 245 of survival for the subject 210 and calculate the loss metric 250.|0127] In some embodiments, the loss metric 250 may correspond to a loss function to quantify the difference between the classification 245 and the label 230 within the training data 205 to optimize ML architecture 150 during training. The loss function can be at least one of Mean Squared Error, Binary Cross-Entropy Loss, Categorical Cross-Entropy Loss, Hinge Loss, or Wasserstein Loss. Using the loss function for the aggregate predictor 175 and the classifier 180, the ML architecture can avoid any issues of mode collapse and vanishing gradients to allow the model trainer 135 to provide more stable training for the ML architecture 150. Therefore, the system 100 can effectively adapt to each subject 210 with relative ease, low computer resources, and stable training for the ML architecture 150.

[0128] The model trainer 135 can compare the classification 245 with the label 230 of the training data 205 to generate the loss metric 250 according to a delta from a threshold. The threshold can indicate an optimal value for the comparison. When the comparison is at the threshold, the model trainer 135 may not generate a loss metric 250 for the ML architecture 150. When the comparison deviates from the threshold (e.g., above or below), the delta can be used to generate the loss metric 250. Therefore, a high delta (e.g., large deviation from threshold) canAtty. Dkt. No.: 115872-3344 cause the model trainer 135 to generate a high loss metric 250. Conversely, a low delta (e.g., small deviation from the threshold) can cause the model trainer to generate a low loss metric 250.

[0129] The model trainer 135 can update, further train, or otherwise tune the at least one of a plurality of weights in the ML architecture 150 using the loss metric 250. In some embodiments, the model trainer 135 can adjust the plurality of weight so the ML architecture 150 to generate more accurate classifications 245. For example, the model trainer 135 can use the loss metric 250 to compute gradients for the direction and magnitude of parameter adjustments to minimize the loss metric 250. In another instance, the model trainer 135 can carry out optimization algorithms such as stochastic gradient descent or RMS prop to update model parameters iteratively. In some implementations, the direction and step size of the parameter update according to the gradients and the optimization algorithm. The model trainer 135 can update or otherwise tune each of the encoders within the ML architecture using the loss metric 250.

[0130] Referring now to FIG. 22B, depicted is a block diagram of a process 255 of training a clustering model 190 to determine an assignment for each cluster within a feature space using the plurality of embeddings (e.g., embeddings 235, embeddings 235’, embeddings 235”). Under the process 255, the model trainer 135 can execute the clustering model 190 using one or more of the embeddings 235, 235’, and 235”. The training of the clustering model 190 can be done in conjunction with the training of the remainder of the ML architecture 150 or after completion of the training of the remainder of the ML architecture 150. The model trainer 135 can provide, input, or otherwise feed each of the embeddings 235 (e.g., embeddings 235, embeddings 235’, embeddings 235”) to the clustering model 190. The model trainer 135 can determine or generate a plurality of embeddings 260 based on each of the embeddings (e.g., embeddings 235, embeddings 235’, embeddings 235”) to input into the clustering model 190. The clustering model 190 can include or define a feature space 265. The feature space 265 can be an / / -dimensional space in which each embedding 260 derived from the WBC images 220, RBC images 220, and the non-image dataset 225. The feature space 265 can define or include a plurality of regions 270A-N (hereinafter generally referred to as regions 270). Each region 270 can correspond to a portion of the feature space 265 of the clustering model 190. With the execution,Atty. Dkt. No.: 115872-3344 the model trainer 135 can determine or identify a cluster assignment 280A-N (hereinafter generally referred to as cluster assignments 280) for each embedding 260 in the clustering model 190. The cluster assignment 280 can define or identify which cluster 275 a given set of embeddings 260 is assigned to within the feature space 265 defined by the clustering model 190. The model trainer 135 can iterate over the embeddings 260 to determine the respective cluster assignment 280.

[0131] During training, the model trainer 135 may calculate, generate, or otherwise determine at least one loss metric 285 to update the clustering model 190. The loss metric 285 may be determined based on the cluster assignments 280 for the embeddings 260 in the feature space 265. Using the cluster assignment 280, the model trainer 135 can determine the loss metric 285 for each of the clusters 275. The loss metric 285 can be used to update the regions 270 defining the clusters 275 in the feature space 265. For instance, the loss metric 285 can define an update to a centroid for a given region 270 for a corresponding cluster 275 in the feature space 265. The loss metric 285 can be a combination (e.g., mean) of the values in the embeddings 260 in the given region 270 for the cluster 275. In accordance with the loss metric 285, the model trainer 135 can modify, adjust, or otherwise update the regions 270 of the clustering model 190. With the updating of the regions 270, the model trainer 135 can modify, alter, or otherwise update the cluster assignments 280 of the embeddings 260. With the re-assignment of the centroids, the training and application process may be repeated as described above upon convergence.

[0132] FIG. 23A is a block diagram of a process 300 determining values indicating a probability of conditions of subjects. The process 300 can include or correspond to operations performed in the system 100. Under the process 300, the dataset indexer 125 can obtain, retrieve, or receive one or more images 320A-N (sometimes referred to as images 320) of a blood sample 315 obtained from a subject 310 from the imaging device 110. The images 320 can be captured by the imaging device 110 in accordance with any number of imaging techniques, such as light microscopy, digital microscopy, fluorescence microscopy, flow cytometry, scanning electron microscopy, transmission electron microscopy, photographs, among others.Atty. Dkt. No.: 115872-3344[0133| A clinician (e.g., nurse or physician examining the subject 310) can extract, obtain, or otherwise retrieve the blood sample 315 from the subject 310. The blood sample 315 can be extracted from peripheral veins, a central venous catheter, or bone marrow of the subject 310 for analysis. The blood sample 315 may be acquired in accordance with a peripheral blood smear (PBS). Once extracted, the physician can use the imaging device 110 to capture the images 320 of the blood sample 315. The blood sample 315 can include various compositions of the blood, such as, circulating tumor cells, leukemic cells, prostate-specific antigens, CA-125, Carcinoembryonic Antigen, Alpha-fetoprotein, CA 19-9, altered blood cell counts (e g., low red blood cells, low white blood cells) C-reactive Proteins, among other compositional changes (e.g., due to cancer or other medical condition). The blood sample 315 can include RBC, WBC, plasma, platelets, enzymes, minerals, glucose, lipids, serum, blood proteins, among others.

[0134] The subject 310 may be afflicted with or at risk of developing cancer. The cancer can at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases, among others, images 320Upon acquisition of the image 320, the imaging device 110 can transmit the image 320 of the blood sample 315 to the dataset indexer 125. The images 320 may be similar to the images 220 detailed herein.[0135| Concurrently, the administrative device 115 can transmit, send, or otherwise provide a non-image dataset 325 to the dataset indexer 125. The non-image dataset 325 can include characteristics, information, and traits about the subject 310. The traits of the subject 310Atty. Dkt. No.: 115872-3344 can be provided by the subject 310, extracted from an electronic medical record, or documented by a physician. The traits can include symptoms (e.g., fatigue, weakness, bleedings), medical history, family medical history, lab results, physical traits, psychological traits (e.g., personality, coping mechanism, emotional state), behavioral traits (e.g., adherence to treatment, health seeking behavior, communication style), social traits, among other traits. The non-image dataset 325 can include a blood count derived from the blood sample 315 of the subject 310. The blood count can be a complete blood count, RBC count, WBC count, hemoglobin, platelet count, among others. In some embodiments, the blood count may indicate an estimated total blood count in the body of the subject 310. For example, the blood count may be derived as a function of the blood count in the blood sample 215. The non-image dataset 325 can further include a plurality of parameters derived from testing of the blood sample 315. The plurality of parameters (e.g., chemistry panel parameters) can include one or more of: glucose level, calcium level, sodium level, potassium level, chloride level, carbon dioxide level, blood urea nitrogen (BUN), a creatinine level, albumin level, total protein level, alkaline phosphatase (ALP) level, alanine aminotransferase (ALT) level, aspartate aminotransferase (AST) level, anion gap, or bilirubin level, among others. The non- image dataset 325 can include a plurality of physiological measurements of the subject 310. The plurality of physiological measurements can include vital signs, temperature, heart rate, respiratory rate, blood pressure, oxygen saturation, age, height, weight, body mass index (BMI), and body surface area (BSA), among others.[0136J The dataset indexer 125 can receive the images 320 of the blood sample 315 taken from the subject 310 from the imaging device 110. In addition, the dataset indexer 125 can receive the non-image dataset 325 from the administrative device 115 (or another computing this device). With receipt, the dataset indexer 125 may process or parse the images 320 and the non-image dataset 325. In some embodiments, the dataset indexer 125 may partition, section, or otherwise divide a single image 320 into a set of patches forming the images 320. Each patch may correspond to a respective portion of the image 320. For example, the imaging device 110 may originally generate a single image 320 for the blood sample 315, from which the dataset indexer 125 divides to form the plurality of images 320 for analysis.Atty. Dkt. No.: 115872-3344[0137| The image sorter 130 can identify, determine, or otherwise select WBC images 320’A-N (sometimes referred to as WBC images 320’) from the images 320 provided by the dataset indexer 125. The image sorter 130 can analyze each image 320 to identify the WBC images 320’. In operation, the image sorter 130 can obtain each image 320 from the plurality of images 320 from the dataset indexer 125. Upon reception of an image 320, the image sorter 130 can preprocess the image. While preprocessing, the image sorter 130 can remove noise, distortions, artifacts, or stains associated with the image 320 and perform contrast enhancement (e.g., adaptive contrast stretching, histogram equalization) between the cells of the blood sample 315 and the background of the image 320. Once each image 320 is preprocessed, the image sorter 130 can execute segmentation to identify at least one portion, region, or object of the image 320 to separate the WBCs from the background and non-similar cells (e.g., a WBC from a group of RBCs). The image sorter 130 can use thresholding, algorithms (e.g., watershed algorithm), or edge detection, among other methods to execute segmentation.

[0138] Continuing on, the image sorter 130 can extract features from each cell in the image 320 to distinguish the WBCs from the RBCs. The features can include, but are not limited to, shape, size, color, intensity, nucleus to cytoplasm ratio, granularity, among other distinguishing features. In some embodiments, the image sorter 130 can use a ML algorithm to classify the WBCs. The ML algorithm can be, for example, support vector machine (SVM), Random Forest, K-Nearest Neighbor (KNN), deep learning CNN, another other algorithms. Once classified, the image sorter 130 can identify the WBC image 320’ as an image 320 that mostly (e.g., at least 50- 70%) contains WBCs from the images 320 of the blood sample 315.

[0139] Concurrently, the image sorter 130 can identify, determine, or otherwise select RBC images 320’A-N (sometimes referred to as RBC images 320”). The image sorter 130 can analyze each image 320 to identify the RBC images 320”. In operation, the image sorter 130 can obtain each image 320 from the plurality of images 320 from the dataset indexer 125. Upon reception of an image 320, the image sorter 130 can preprocess the image. While preprocessing, the image sorter 130 can remove noise, distortions, artifacts, or stains associated with the image 320 and perform contrast enhancement (e.g., adaptive contrast stretching, histogram equalization)Atty. Dkt. No.: 115872-3344 between the cells of the blood sample 315 and the background of the image 320. Once each image 320 is preprocessed, the image sorter 130 can execute segmentation to identify at least one portion, region, or object of the image 320 to separate the RBC images 320” from the background and nonsimilar cells (e.g., an RBC from a group of WBCs). The image sorter 130 can use thresholding, algorithms (e.g., watershed algorithm), or edge detection, among other methods to execute segmentation.

[0140] Continuing on, the image sorter 130 can extract features from each cell in the image 320 to distinguish the RBCs from the WBCs. The features can include, but are not limited to, shape, size, color, intensity, nucleus to cytoplasm ratio, granularity, among other distinguishing features. For example, depending on the staining of the sample, the RBC images 220" may have a redder hue than WBC images 220’. In some embodiments, the image sorter 130 can use a ML algorithm to classify the RBCs. The ML algorithm can be, for example, a support vector machine (SVM), Random Forest, K-Nearest Neighbor (KNN), deep learning CNN, and other algorithms. Once classified, the image sorter 130 can identify the RBC image 320” as an image 320 that mostly (e g., at least 50-70%) contains RBCs from the images 320 of the blood sample 315.[0141 | The model applier 140 can apply, feed, or otherwise input the WBC images 320’ to the ML architecture 150. The model applier 140 can apply, execute, or otherwise provide the ML architecture 150 for the one or more subjects 310. The ML architecture 150 can include the at least one WBC encoder 160, the at least one aggregate predictor 175 and at least one classifier 180, among others. The WBC encoder 160 can ingest, receive, or otherwise obtain the WBC images 320’ from the model applier 140. The ML architecture 150 can include a plurality of parameters for determine the value indicating the probability of condition. In some embodiments, the plurality of parameters can include a plurality of hyperparameters to establish the architecture, design, or configuration of the ML architecture 150 to determine the value indicating the probability of condition.[0142 j Within the ML architecture 150, the plurality of hyperparameters can be arranged across the WBC encoder 160, the RBC encoder 165, the aggregate predictor 175, and the classifierAtty. Dkt. No.: 115872-3344180. The plurality of hyperparameters can include a learning rate, number of epochs, a batch size, a model architecture (e.g., deep learning convolutional neural network, ensemble network, or clustering algorithm, or any combination thereof), and regularization parameters. In some implementations, the plurality of parameters can include a plurality of model parameters. The plurality of model parameters can be variables learned from the training data 205 (i.e., one or more examples) that establish, dictate, or otherwise define a link between the input (e.g., WBC images 320’) and the output (e.g., classification 340). The model parameters can continuously optimize to minimize a loss function of the ML architecture 150 during use. The plurality of model parameters can include target weights or biases (e.g., influence the output of the ML architecture 150 based on accurate classifications 340), loss function (e.g., compound loss), among others. This is not limited to the WBC encoder 160, but the hyperparameters of the ML architecture 150 can be arranged and adjusted for the RBC encoder 165 in a similar manner.|0143] In feeding, the WBC encoder 160 can generate, produce, or otherwise determine a set of WBC embeddings 33OA-N (sometime referred to as embeddings 330). To generate the set of WBC embeddings 330, the WBC encoder 160 can execute dimensionality reduction, feature extraction, or sequence modeling. In some implementations, the WBC encoder 160 can include a plurality of layers, to extract, retrieve, or otherwise obtain the WBC images 320’ from the image sorter 130 and the model applier 140. The plurality of layers can include one or more hidden layers. The one or more hidden layers can transform the WBC images 320’to extract or retrieve one or more representations of the WBC images 320’ . The one or more representations can include the set of WBC embeddings 330.

[0144] Concurrently, the model applier 140 can apply, feed, or otherwise input the RBC images 320” to the ML architecture 150. The model applier 140 can apply, execute, or otherwise employ the ML architecture 150 for the one or more subjects 310. The ML architecture 150 can include the at least one RBC encoder 165, the at least one aggregate predictor 175 and at least one classifier 180, among others. The RBC encoder 165 can ingest, receive, or otherwise obtain the RBC images 320” from the model applier 140. The ML architecture 150 can include a plurality of parameters for determine the value indicating the probability of condition. In someAtty. Dkt. No.: 115872-3344 embodiments, the plurality of parameters can include a plurality of hyperparameters to establish, design, or configuration of the ML architecture 150 to determine the value indicating the probability of condition.|01451 Within the ML architecture 150, the plurality of hyperparameters can be arranged across the RBC encoder 165, the RBC encoder 165, the aggregate predictor 175, and the classifier 180. The plurality of hyperparameters can include a learning rate, number of epochs, a batch size, a model architecture (e.g., deep learning convolutional neural network), and regularization parameters. In some implementations, the plurality of parameters can include a plurality of model parameters. The plurality of model parameters can be variables learned from the training data 205 (i.e., one or more examples) that establish, dictate, or otherwise define a link between the input (e.g., RDC images 320”) and the output (e.g., classification 340). The model parameters can continuously optimize to minimize a loss function of the ML architecture 150. The plural of model parameters can include target weights or biases (e.g., influence the output of the ML architecture 150 based on accurate targets), loss function (e.g., compound loss), among others. This is not limited to the RBC encoder 165, but the hyperparameters of the ML architecture 150 can be arranged and adjusted for the WBC encoder 160 in a similar manner.[0146| In feeding, the RBC encoder 165 can generate, produce, or otherwise determine a set of RBC embeddings 330’A-N (sometime referred to as embeddings 330’). To generate the set of RBC embeddings 330’, the RBC encoder 165 can execute dimensionality reduction, feature extraction, or sequence modeling. In some implementations, the RBC encoder 165 can include a plurality of layers to extract, retrieve, or otherwise obtain the RBC images 320” from the image sorter 130 and the model applier 140. The plurality of layers can include one or more hidden layers. The one or more hidden layers can transform the RBC images 320” to extract or retrieve one or more representations of the RBC images 320”. The one or more representations can include the set of RBC embeddings 330’.[0147| The model applier 140 can apply, feed, or otherwise input the non-image dataset 325 to the feature encoder 170. The model applier 140 can apply, execute, or otherwise use theAtty. Dkt. No.: 115872-3344ML architecture 150 on the non-image dataset 325 for the subject 310. The feature encoder 170 can ingest, receive, or otherwise obtain the non-image dataset 325 from the model applier 140. From feeding, the feature encoder 170 can generate, establish or otherwise determine a set of feature embeddings 330” A-N (referred to as embedding 330” herein). The feature encoder 170 (similar to the RBC encoder 165 and the WBC encoder 160) can execute one or more of dimensionality reduction, feature extraction, on the data within the non-image dataset 325. The model applier 140 can transform the non-image dataset 325 into a feature vector prior to ingestion by the feature encoder 170. In this manner, the feature encoder 170 can execute the functions (e.g., feature extraction) on the feature vector. The feature encoder 170 can include a plurality of layers to extract, retrieve, or otherwise obtain the non-image dataset 325 from the model applier 140 and the image sorter 130. The plurality of layers can include an input layer, hidden layers (e.g., a fully connected layer, batch normalization layer, activation layer), and an output layer. The input layer can receive the feature vector or transform the non-image dataset 325 into the feature vector. The hidden layers can process the feature vector by applying a linear transformation and execute anon- linear function on each aspect of the non-image dataset 325 (e.g., traits, parameters) to generate intermediary output values. Each of the intermediary output values can indicate patterns or connection for the output layers to establish the final embeddings 330. The output layer can generate each of the embeddings 330” using the intermediary output values.

[0148] Upon generation of the embeddings 330, the WBC encoder 160 can transmit, feed, or otherwise input the embeddings 330 into the aggregate predictor 175. In a similar manner, upon generation of the embeddings 330’, the RBC encoder 165 can transmit, feed, or otherwise input the embeddings 330’ to the aggregate predictor 175. In a similar manner, the feature encoder 170 can provide, feed, or otherwise input the embeddings 330” to the aggregate predictor 175. The aggregate predictor 175 can receive, ingest, or otherwise obtain the embeddings 330 and the embeddings 330’. The aggregate predictor 175 can receive, ingest, or otherwise obtain at least one or more of the embeddings 330, the embeddings 330’, and the embeddings 330”. The aggregate predictor 175 can generate, determine, or otherwise predict a value 335 indicating a probability of mortality for the subject 310 based on at least one of the embeddings 330, theAtty. Dkt. No.: 115872-3344 embeddings 330’, or the embeddings 330”. The aggregate predictor 175 can generate, determine, or otherwise predict a value 335 indicating a probability of a condition of the plurality of conditions associated with the mortality for the subject 310 based on the embeddings 330, the embeddings 330’, and the embeddings 330”. The aggregate predictor 175 can execute, for example, bagging (e g., random forests), boosting (e.g., gradient boosting), or stacking to determine the value 335 indicating the probability of mortality.

[0149] From the execution, the aggregate predictor 175 can generate the value 335 based on the embeddings 330 and the embeddings 330’. The value 335 may indicate the probability of mortality for the subject 310 for the given time interval. The probability of mortality can indicate a likelihood that the subject 310 can die based on the condition of the RBC and WBC within the RBC images 320” and the WBC images 320’ . The time interval may be relative to the time of the extraction of the blood sample 215. The time interval can be any range between 6 hours to 6 months, for example, at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 60 days, 90 days, 120 days, or 180 days, among other time intervals. In some embodiments, with the determination of the probability for each of the RBC and WBC images, the aggregate predictor 175 can combine the predictions to determine the value 335 indicating the probability of mortality. To determine the value, the aggregate predictor 175 can implement averaging, majority voting, weighted averaging, meta-model in stacking, among other methods to determine the value 335.

[0150] In some embodiments, the aggregate predictor 175 can generate the value 335 indicating a probability of the presence (or absence) of a condition of the plurality of conditions associated with the mortality for the subject 310 based on at least one of the embeddings 330, the embeddings 330’, and the embeddings 330”. The probability of the condition can indicate a likelihood of the presence of the condition in the subject 310. The probability of the condition can be independent of time. In some embodiments, the aggregate predictor 175 can generate a plurality of values 335 corresponding to the plurality of conditions based on at least one of the embeddings 330’, the embeddings 330’, and the embeddings 330’”. For each condition, the value 335 can indicate the probability of the presence (or absence) of the condition in the subject 310. TheAtty. Dkt. No.: 115872-3344 aggregate predictor 175 can feed, input, or otherwise provide the value 335 indicating mortality to the classifier 180. In some embodiments, the aggregate predictor 175 can determine or generate the value 335 to indicate a likelihood of clinical decision to be carried out for the subject 210 based on the embeddings 330, the embeddings 330, and the embeddings 330”. For each candidate clinical decision, the aggregate predictor 175 can determine or generate the value 335 to indicate likelihood of taking the candidate clinical decision.[0151 In some embodiments, the model applier 140 can feed, input or otherwise provide the non-image dataset 325 to the classifier 180. The classifier 180 can include one or more input layers that receives the value 335 indicating the probability of condition and the non-image dataset 325 to provide the value 335 and the non-image dataset 325 to the subsequent layers of the classifier 180. In some embodiments, the classifier 180 can execute, for example, bagging (e.g., random forests), boosting (e.g., gradient boosting), or stacking to determine the value 335 indicating the probability of condition for the time interval. In some embodiments, the classifier 180 can include one or more hidden layers (e.g., fully connected layers, convolutional layers, recurrent layers) to learn patterns and representations of the value 335 with the non-image dataset 325 based on an activation function (e.g., ReLU, Sigmoid, Tanh). For instance, the one or more hidden layers can form a pattern based on the value 335 and a trait of the subject 310 within the non-image dataset 325.

[0152] Using the patterns and representations, the output layer of the classifier 180 can provide at least one classification 340 for the subject 310. The output layer can be at least one of binary classification or multi class classification. The classification 340 for the subject 310 can be at least one of mortality or survival for the subject 310 in the given time interval based on the inputs of the value 335 and the non-image dataset 325. In some embodiments, in processing the inputs, the classifier 180 may calculate, generate, or otherwise determine an intermediary value. The value may indicate a probability or mortality for the subject 305 based on the value 335 calculated from the images 320’ and 320” and the non-image dataset 325. The classifier 180 may compare the intermediary value to a threshold. If the value satisfies (e.g., greater than or equal to) the threshold, the classifier 180 may generate the classification 340 to indicate that the subject 310Atty. Dkt. No.: 115872-3344 is at risk of dying for at least the given time interval. Otherwise, if the value does not satisfy (e.g., less than) the threshold, the classifier 180 may generate the classification 340 to indicate survival for at least the given time interval. When the classification 340 corresponds to mortality, the subject 310 can be at risk of dying because of the cancer and / or condition. However, when the classification 340 corresponds to survival, the subject 310 is not at risk of dying to the cancer and / or condition. In some embodiments, the classification 340 may indicate or identify the clinical decision to be applied to the subject 310.

[0153] In some embodiments, the classifier 180 can generate the classification 345 using the value 335 indicating the probability of the condition associated with the mortality of the subject 310. The classifier 180 can use the value 335 to generate the classification indicating an absence or a presence of the condition. The classifier 180 can generate an intermediary value using the value 335. The intermediary value can be a binary, flag, or marker that indicates the respective condition impacting the blood sample 315 of the subject 310. The classifier 180 may compare the intermediary value to a threshold. If the value 335 satisfies (e.g., greater than or equal to) the threshold, the classifier 180 can generate the classification 345 to indicate the presence of the condition. On the other hand, if the value 335 does not satisfy (e.g., less than) the threshold, the classifier 180 can generate the classification 345 to indicate the absence of the condition. The classifier 180 can iterate through the set of values 335 for the various conditions.

[0154] In some embodiments, the classifier 180 can generate the classification 245 to indicate which clinical decision is to be carried out for the subject 210 based on the value 240. The classifier 180 can generate an intermediary value using the value 240. The intermediary value can be a binary, flag, or marker that indicates the respective clinical decision to be taken (e.g., to effectuate the proper procedure or action to not take on the subject 210). The classifier 180 may compare the intermediary value to a threshold. If the value 240 satisfies (e.g., greater than or equal to) threshold, the classifier 180 can generate the classification 245 to indicate the selection of the clinical decision. On the other hand, if the value 240 does not satisfy (e.g., less than) threshold, the classifier 180 can generate the classification 245 to indicate the exclusion of the clinical decision.Atty. Dkt. No.: 115872-3344[0155| FIG. 23B depicts a process 400 of determining an assignment using a clustering model. Under the process 400, the model applier 140 can execute the clustering model 190 using one or more of the embeddings 330, 330’, and 330”. The model applier 140 can provide, input, or otherwise feed each of the embeddings 330 (e.g., embeddings 330, embeddings 330’, embeddings 330”) to the clustering model 190. The model applier 140 can determine or generate a plurality of embeddings 405 based on each of the embeddings (e.g., embeddings 330, embeddings 330’, embeddings 330”) to input into the clustering model 190. The clustering model 190 can include or define a feature space 410. The feature space 410 can be an w-dimensional space in which each embedding 405 derived from the WBC images 320, RBC images 320, and the non-image dataset 325. The feature space 410 can define or include a plurality of regions 420A-N (hereinafter generally referred to as regions 420). Each region 420 can correspond to a portion of the feature space 410 of the clustering model 190. With the execution, the model applier 140 can determine or identify a cluster assignment 425 A-N (hereinafter generally referred to as cluster assignments 425) for each embedding 405 in the clustering model 190. The cluster assignment 380 can define or identify which cluster 415 a given set of embeddings 405 is assigned to within the feature space 410 defined by the clustering model 190. The model applier 140 can iterate over the embeddings 405 to determine the respective cluster assignment 380.10156) For each cluster assignment 425 (or cluster 415), the model applier 140 can calculate, generate, or otherwise determine a respective set of parameters 430A-N (hereinafter generally referred to as parameters 430). Each embedding 405 in the feature space 410 can correspond to a respective subject 310 or associated data (e.g., the images 320 and non-image dataset 325). The parameters 430 can define or specify characteristics of a given cluster 415. For instance, the parameters 430 can identify an average anion gap of the blood samples and an average ALP level. To determine, the model applier 140 can select or identify the parameters (e.g., chemistry panel parameters and physiological characters) from the non-image dataset 325 associated with the embeddings 405 assigned to the cluster 415. With the identification, the model applier 140 can determine a combination (e.g., a mean, a standard deviation, or otherwise statistical measure) of the values of the parameters across the embeddings 405 in the cluster 415. Based onAtty. Dkt. No.: 115872-3344 the parameters across the embeddings 405 in the cluster 415, the model applier 104 can determine the respective set of parameters 430 for the cluster assignment 425 (or cluster 415). The model applier 140 can iterate over the clusters 415 (or embeddings 405) to determine the set of parameters 430 for each cluster 415 (or embedding 405). The parameters 430 can indicate the contributory causes for the value 335 or the classification 340.[0157J Referring now to FIG. 24, depicted is a block diagram of a process 500 of providing an output 505 to an administrative device. The process 500 can include or correspond to operations performed in the system 100. Under the process 500, the output evaluator 145 can obtain the classification 340 from the ML architecture 150 and the assignments 425 from the clustering model 190. Upon receipt, the output evaluator 145 can store, house, or otherwise maintain an association between the subject 310 and the classification 340 (as well as the value 335, the parameters 430, the assignments 425, the classification 340 and the clinical decision). The association can be a link, a map, or a connection between the subject 310 and the classification 340. To store the association, the output evaluator 145 can generate one or more data structures. The one or more data structures can include an array, a linked list, a stack, a tree, a hash table, among others. For example, the data structure can be a hash table where the subject 310 is the key to the hash table and the classification 340 is the value of the hash table. The hash table can include a plurality of keys (i.e., for each subject 310) mapped to a plurality of values (i.e., classification 340 of the subject 310). In another example, the data structure can be a plurality of linked lists. A first linked list can correspond to a first subject 310. Each node in the first linked list corresponds to the classification 340 of the first subject 310. Concurrently, a second linked list can correspond to a second subject 310. Each node in the second linked list corresponds to the classification 340 of the second subject 310. In some implementations, the output evaluator 145 can store an association between the WBC images 320, RBC images 320, and the classification 340.[0158| The output evaluator 145 can generate, determine, or otherwise provide an output 505 according to the classification 340 and the assignments 425 to the administrative device 115. The output 505 can include, for example, the values 335, classification 340, the assignments 425, or the parameters 430 for presentation to a clinician, subject 310, lab technician, or a physicianAtty. Dkt. No.: 115872-3344 through the administrative device 115. In some embodiments, the output evaluator 145 can generate, identify, or otherwise determine a contributory factor for the value 335 or the classification 340 using the set of parameters 430. To determine, the output evaluator 145 can select or identify a value of at least one parameter 430 outside an expected range of values. The expected range may be pre-defined or determined across a cohort of subjects. If the value of at least one parameter 430 is identified as outside the expected range of values, the output evaluator 145 can identify the associated parameter 430 as a contributory factor. For instance, the value of the parameter 430 for heart rate is outside the expected range, the output evaluator 145 can identify the parameter 430 as a contributory factor for the likelihood of morality or the condition as indicated in the value 335. Conversely, if the value of at least one parameter 430 is identified as within the expected range of values, the output evaluator 145 can identify the associated parameter 430 as a non-contributory factor, use.|0159] In some embodiments, using the values 335, the output evaluator 145 can determine or identify a relative score 510 indicating a difference between the subject 310 and the plurality of subjects. The difference can indicate or correspond to a survival probability, a 24-hour risk of mortality, a probability of contracting a condition that indicates mortality of the subject 310. The relative score 510 can be based on the value 335 for the subject 310 (e.g., indicating mortality or another condition) versus the value for subjects in another cohort (e.g., assigned to inpatient group or an intensive care unit (ICU) group). In some embodiments, the relative score 510 can be a function of the values 335, the parameters 430, the assignments 425, or the classification 340, among others. In some embodiments, the output evaluator 145 can identify, determine, or otherwise indicate a category from a set of categories for the subject 310 based on the relative score 510 and a score for the category. The categories can include, for example, one or more of ICU Medium, ICU Low, Inpatient Medium, ER High, ER Medium, ER Low, Trauma Room High, Trauma Room Medium, Trauma Room Low, Step-Down Unit High, Step-Down Unit Low, Surgical Recovery High, Surgical Recovery Low, Oncology Ward High, Oncology Ward Low, Infectious Disease Unit High, and General Ward Low, among other categories. The score for the category can indicate the category by comparing the relative risk of mortality to the relativeAtty. Dkt. No.: 115872-3344 score of 510 for each of the subjects 310 within a unit. The relative score 510 can be compared against predetermined thresholds. For example, a high relative risk score above a certain value (e.g., >2.0) can place a patient into “ICU High,” while lower scores could fall into “ICU Medium” or “ICU Low.” These thresholds can be defined statistically based on historical patient outcomes in each location, allowing category assignments across the institution.[0160J The output 505 can be, for example, a notification for a clinician to examine the subject 310, a notification to administer an intervention within the time interval, or a notification for the subject 310 to request for medical attention. In some instances, the classification 340 can indicate that the subject 310 can survive for the time interval, in response to the value 335 satisfying the threshold. Accordingly, the output evaluator 145 can provide the output 505 as a notification to the administrative device 115 of a clinician to indicate that the subject 310 can survive for the time interval. In another instance, the classification 340 can indicate that the subject 310 is at risk of dying, in response to the value 335 satisfying the threshold. In some embodiments, the output evaluator 145 can provide the output 505 as a notification to the administrative device of a physician indicating that the subject 310 is at risk of dying within the time interval. In some embodiments, the output evaluator 145 can provide the output 505 to indicate the clinical decision identified in the classification 340.[016.1] The output 505 may indicate the clinical decision (e.g., by the institution or clinician in care of the subject 310) to be carried out as determined using the ML architecture 150. The clinical decision may include, for example, any one or more of the following: (1) admission level; (2) monitoring frequency; (3) diagnostic testing; (4) medication adjustment; (5) surgical timing; (6) transport decisions; (7) nutritional support; (8) ventilation support; (9) discharge planning; (10) fluid management; (11) blood transfusions; (12) physical activity and rehabilitation; (13) consultations; (14) code status discussion; (15) sepsis and infection management; (16) pain management; (17) fluid restriction; (18) medication route; (19) sedation and anesthesia level; (20) anti coagulation management; (21) vital signs review frequency (22) electrolyte management; (23) infection control precautions; (24) blood pressure management protocol; (25) family visitation; (26) ambulation restrictions; (27) temperature management; (28) electrocardiogram (ECG)Atty. Dkt. No.: 115872-3344 monitoring; (29) risk of pressure injuries; (30) end-of-life or palliative care initiatives; (31) ICU bed allocation; or (32) emergency dialysis initiation, among others, as detailed herein. The output 505 can be used to implement or suggest one or more of the parameters for the subject 310 to reduce the risk associated with the category. The clinician can apply the parameters to the subject 310 upon display of the user interface 515.[0162 j Upon receipt of the output 505, the administrative device 115 may present, render, or otherwise display the output 505 for interpretation by a clinician on a user interface 515. The user interface 515 can include a plurality of user interface elements that can be configured to display, render, or otherwise display the output 505. In some instances, the user interface 515 can be generated by instructions within the output 505. The user interface 515 can include, for example, one or more of (i) a category for the first subject, (ii) a relative score between the first subject and a plurality of subjects, (iii) the value indicating the probability of mortality for the first subject, (iv) the set of parameters for a condition, (v) the condition of the subject 310. Once the output 505 is presented, the administrative device 115 can automatically obtain, retrieve, or otherwise access the non-image dataset 325 associated with the subject 310. The administrative device 115 can extract the non-image dataset 325 from the database 155 based on the respective subject 310. Upon retrieval of the non-image dataset 325, the administrative device 115 can use the output 505 to automatically update the non-image dataset with the classification 340 for the subject 310. In this manner, the administrative device 115 can continuously update the non-image dataset 325 at multiple iterations of blood sample analysis for a subject 310.[01631 Referring now to FIG. 25, depicted is an example of the user interface 515 displayed on the administrative device. The user interface 515 can be a clinician-facing interface designed to present the peripheral blood smear-based risk predictions in an interpretable and actionable format. The user interface 515 can include user interface elements to present morphology findings, laboratory values, and risk scores to support early identification of patients at high risk for clinical deterioration. The left-hand “Risk Thermometer” can include an indicator of risk to provide a rapid visual assessment of patient acuity, with font size and style adjusted to emphasize urgent cases. Each row can include identifiers such as medical recordAtty. Dkt. No.: 115872-3344 number, name, date and time of sample collection, and the originating clinical location (e.g., ER, ICU, inpatient ward). The Risk Category column conveys the classification (e g., “ICU High,” “Inpatient Low”) alongside calculated 24-hour relative risk of mortality compared to other patients in similar wards, and a projected 5-day survival probability. The Abnormalities column lists specific morphological and biochemical changes detected (e.g., schistocytes, echinocytes, elevated band neutrophils, abnormal anion gap, or low sodium) that contributes to the risk score. The Syndrome Risk column flags likely clinical syndromes, including disseminated intravascular coagulation (DIC), metabolic disturbances, or bacterial infection / sepsis, among others. In the user interface 515, the columns may be sortable and searchable, allowing the user (e.g., clinician) to filter and prioritize cases in real time. The user interface 515 can also provide direct access to the image data or non-image data.

[0164] In this manner, the data processing system 105 can use the ML architecture 150 to process data from multiple modalities (e.g., image and non-image data) to perform an assessment of the mortality of the subject. The use of data from multiple, different modalities used to encode the parameters of the ML architecture 150 can provide for enhanced accuracy and greater precision, reducing false positives or negatives. From a clinical perspective, the different modalities may also provide for better insights as to the mortality risk of the subject to make personalized decisions regarding the medical care of the subject. Since the ML architecture 150 may be able to achieve accurate assessments in near-real-time, the data processing system 105 can generate automated alerts when the subject is classified at risk of dying to provide to the subject or a clinician for the subject to seek immediate medical assistance. From a computer resource perspective, the use of the integrated ML architecture 150 to process the multimodal data can allow for feature sharing across modalities, leveraged correlated features, reducing the reliance for redundant calculations. The reduction in redundant calculations for each of the modalities can save computing resources (e.g., in terms of processing and memory) as well as time (e.g., from executing the individual models individually). The use of the integrated ML architecture 150 can allow for streamlined processing, minimizing repetitive cleaning or transformation tasks. The data processing system 105 is thus able to provide results quicker than other uni-modal approaches.Atty. Dkt. No.: 115872-3344[0165| FIG. 26 is a flow diagram of a method of determining values indicating probabilities of mortality of subjects using blood samples. The method 600 can be implemented or performed by any components detailed herein, such as system 100 or system 800. Under the method 600, a computing system can receive a first plurality of images of a first blood sample having first white blood cells (WBCs) and first red blood cells (RBCs) obtained from a first subject at a first time (605). The computing system can identify, from the first plurality of images of the first blood sample, (i) a first set of WBC images corresponding to the first WBCs and (ii) a first set of RBC images corresponding to the first RBCs (610). The computing system can apply the first set of WBC images and the first set of RBC images to a machine learning (ML) architecture (615). The computing system can determine based on applying the images to the ML architecture, a value indicating a probability of condition for the first subject at the time interval relative to the first time (620). The computing system can generate a classification of the first subject as one of mortality or survival in accordance with the value indicating the probability of condition (625). The computing system can provide an output based on the classification (630).[Q166| FIG. 27 is a flow diagram of method of training machine learning (ML) architectures to determine values indicating expected mortality of subjects using blood samples. The method 700 can be implemented or performed by any components detailed herein, such as system 100 or system 800. Under the method 700, a computing system can retrieve a training dataset including a plurality of examples (705). The computing system can identify, from the plurality of images of the blood sample of at least one example of the plurality of examples in the training dataset (710). The computing system can apply the set of WBC images and the set of RBC images to a machine learning (ML) architecture (715). The computing system can determine, based on applying the images to the ML architecture, a value indicating a probability of condition for the first subject at the time interval relative to the first time (720). The computing system can generate a classification of the subject as one of mortality or survival in accordance with the value indicating the probability of condition (725). The computing system can compare the classification of the subject generated in accordance with the value determined by the ML architecture with theAtty. Dkt. No.: 115872-3344 label of the training dataset (730). The computing system can update at least one of the plurality of weights in the ML architecture based on comparing the classification and the label (735).C. Network and Computing Environment

[0167] Various operations described herein can be implemented on computer systems. FIG. 28 shows a simplified block diagram of a representative server system 800, computing system 814, and network 826 usable to implement certain embodiments of the present disclosure. In various embodiments, server system 800 or similar systems can implement services or servers described herein or portions thereof. Computing system 814 or similar systems can implement clients described herein. The system 100 described herein can be similar to the server system 800. Server system 800 can have a modular design that incorporates a number of modules 802 (e.g., blades in a blade server embodiment); while two modules 802 are shown, any number can be provided. Each module 802 can include processing unit(s) 804 and local storage 806.[0168| Processing unit(s) 804 can include a single processor, which can have one or more cores, or multiple processors. In some embodiments, processing unit(s) 804 can include a general- purpose primary processor as well as one or more special-purpose co-processors such as graphics processors, digital signal processors, or the like. In some embodiments, some, or all processing units 804 can be implemented using customized circuits, such as application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs). In some embodiments, such integrated circuits execute instructions that are stored on the circuit itself. In other embodiments, processing unit(s) 804 can execute instructions stored in local storage 806. Any type of processors in any combination can be included in processing unit(s) 804.

[0169] Local storage 806 can include volatile storage media (e.g., DRAM, SRAM, SDRAM, or the like) and / or non-volatile storage media (e.g., magnetic, or optical disk, flash memory, or the like). Storage media incorporated in local storage 806 can be fixed, removable, or upgradeable as desired. Local storage 806 can be physically or logically divided into various subunits such as a system memory, a read-only memory (ROM), and a permanent storage device.Atty. Dkt. No.: 115872-3344The system memory can be a read-and-write memory device or a volatile read-and-write memory, such as dynamic random-access memory. The system memory can store some or all of the instructions and data that processing unit(s) 804 need at runtime. The ROM can store static data and instructions that are needed by processing unit(s) 804. The permanent storage device can be a non-volatile read-and-write memory device that can store instructions and data even when module 802 is powered down. The term “storage medium” as used herein includes any medium in which data can be stored indefinitely (subject to overwriting, electrical disturbance, power loss, or the like) and does not include carrier waves and transitory electronic signals propagating wirelessly or over wired connections.

[0170] In some embodiments, local storage 806 can store one or more software programs to be executed by processing unit(s) 804, such as an operating system and / or programs implementing various server functions such as functions of the system 100 or any other system described herein, or any other server(s) associated with system 100 or any other system described herein.

[0171] “Software” refers generally to sequences of instructions that, when executed by processing unit(s) 804, cause server system 800 (or portions thereof) to perform various operations, thus defining one or more specific machine embodiments that execute and perform the operations of the software programs. The instructions can be stored as firmware residing in read-only memory and / or program code stored in non-volatile storage media that can be read into volatile working memory for execution by processing unit(s) 804. Software can be implemented as a single program or a collection of separate programs or program modules that interact as desired. From local storage 806 (or non-local storage described below), processing unit(s) 804 can retrieve program instructions to execute and data to process in order to execute various operations described above.

[0172] In some server systems 800, multiple modules 802 can be interconnected via a bus or other interconnect 808, forming a local area network that supports communication betweenAtty. Dkt. No.: 115872-3344 modules 802 and other components of server system 800. Interconnect 808 can be implemented using various technologies, including server racks, hubs, routers, etc.

[0173] A wide area network (WAN) interface 810 can provide data communication capability between the local area network (e.g., through the interconnect 808) and the network 826, such as the Internet. Other technologies can be used to communicatively couple the server system 800 with the network 826, including wired (e.g., Ethernet, IEEE 802.3 standards) and / or wireless technologies (e.g., Wi-Fi, IEEE 802.11 standards).

[0174] In some embodiments, local storage 806 is intended to provide working memory for processing unit(s) 804, providing fast access to programs and / or data to be processed while reducing traffic on interconnect 808. Storage for larger quantities of data can be provided on the local area network by one or more mass storage 812 that can be connected to interconnect 808. Mass storage 812 can be based on magnetic, optical, semiconductor, or other data storage media. Direct attached storage, storage area networks, network-attached storage, and the like can be used. Any data stores or other collections of data described herein as being produced, consumed, or maintained by a service or server can be stored in mass storage 812. In some embodiments, additional data storage resources may be accessible via WAN interface 810 (potentially with increased latency).

[0175] Server system 800 can operate in response to requests received via WAN interface810. For example, one of modules 802 can implement a supervisory function and assign discrete tasks to other modules 802 in response to received requests. Work allocation techniques can be used. As requests are processed, results can be returned to the requester via WAN interface 810. Such operation can generally be automated. Further, in some embodiments, WAN interface 810 can connect multiple server systems 800 to each other, providing scalable systems capable of managing high volumes of activity. Other techniques for managing server systems and server farms (collections of server systems that cooperate) can be used, including dynamic resource allocation and reallocation.Atty. Dkt. No.: 115872-3344[0176| Server system 800 can interact with various user-owned or user-operated devices via a wide-area network such as the Internet. An example of a user-operated device is shown in FIG. 28 as computing system 814. Computing system 814 can be implemented, for example, as a consumer device such as a smartphone, other mobile phone, tablet computer, wearable computing device (e.g., smart watch, eyeglasses), desktop computer, laptop computer, and so on.[0177| For example, computing system 814 can communicate via WAN interface 810. Computing system 814 can include computer components such as processing unit(s) 816, storage device 818, network interface 820, user input 822, and user output 824. Computing system 814 can be a computing device implemented in a variety of form factors, such as a desktop computer, laptop computer, tablet computer, smartphone, other mobile computing device, wearable computing device, or the like.[0178| Processing unit 816 and storage device 818 can be similar to processing unit(s) 804 and local storage 806 described above. Suitable devices can be selected based on the demands to be placed on computing system 814. For example, computing system 814 can be implemented as a “thin” client with limited processing capability or as a high-powered computing device. Computing system 814 can be provisioned with program code executable by processing unit(s) 816 to enable various interactions with server system 800.[0179| Network interface 820 can provide a connection to the network 826, such as a wide area network (e.g., the Internet) to which WAN interface 810 of server system 800 is also connected. In various embodiments, network interface 820 can include a wired interface (e.g., Ethernet) and / or a wireless interface implementing various RF data communication standards such as Wi-Fi, Bluetooth, or cellular data network standards (e.g., 3G, 4G, LTE, etc.).[0180| User input 822 can include any device (or devices) via which a user can provide signals to computing system 814; computing system 814 can interpret the signals as indicative of particular user requests or information. In various embodiments, user input 822 can include anyAtty. Dkt. No.: 115872-3344 or all of a keyboard, touch pad, touch screen, mouse, or other pointing device, scroll wheel, click wheel, dial, button, switch, keypad, microphone, and so on.

[0181] User output 824 can include any device via which computing system 814 can provide information to a user. For example, user output 824 can include display-to-display images generated by or delivered to computing system 814. The display can incorporate various image generation technologies, e.g., a liquid crystal display (LCD), light-emitting diode (LED) display including organic light-emitting diodes (OLED), projection system, cathode ray tube (CRT), or the like, together with supporting electronics (e g., digital-to-analog or analog-to-digital converters, signal processors, or the like). Some embodiments can include a device such as a touchscreen that function as both input and output device. In some embodiments, other user outputs 824 can be provided in addition to or instead of a display. Examples include indicator lights, speakers, tactile “display” devices, printers, and so on.

[0182] Some embodiments include electronic components, such as microprocessors, storage, and memory that store computer program instructions in a computer readable storage medium. Many of the features described in this specification can be implemented as processes that are specified as a set of program instructions encoded on a computer readable storage medium. When one or more processing units execute these program instructions, they cause the processing unit(s) to perform various operations indicated in the program instructions. Examples of program instructions or computer code include machine code, such as is produced by a compiler, and files including higher-level code that are executed by a computer, an electronic component, or a microprocessor using an interpreter. Through suitable programming, processing unit(s) 804 and 816 can provide various functionality for server system 800 and computing system 814, including any of the functionality described herein as being performed by a server or client, or other functionality.

[0183] It will be appreciated that server system 800 and client computing system 814 are illustrative and that variations and modifications are possible. Computer systems used in connection with embodiments of the present disclosure can have other capabilities not specificallyAtty. Dkt. No.: 115872-3344 described here. Further, while server system 800 and client computing system 814 are described with reference to particular blocks, it is to be understood that these blocks are defined for convenience of description and are not intended to imply a particular physical arrangement of component parts. For instance, different blocks can be, but need not be, located in the same facility, in the same server rack, or on the same motherboard. Further, the blocks need not correspond to physically distinct components. Blocks can be configured to perform various operations, e.g., by programming a processor or providing appropriate control circuitry, and various blocks might or might not be reconfigurable depending on how the initial configuration is obtained. Embodiments of the present disclosure can be realized in a variety of apparatus including electronic devices implemented using any combination of circuitry and software.[0184| While the disclosure has been described with respect to specific embodiments, one skilled in the art will recognize that numerous modifications are possible. Embodiments of the disclosure can be realized using a variety of computer systems and communication technologies, including but not limited to specific examples described herein. Embodiments of the present disclosure can be realized using any combination of dedicated components and / or programmable processors and / or other programmable devices. The various processes described herein can be implemented on the same processor or different processors in any combination. Where components are described as being configured to perform certain operations, such configuration can be accomplished, e.g., by designing electronic circuits to perform the operation, by programming programmable electronic circuits (such as microprocessors) to perform the operation, or any combination thereof. Further, while the embodiments described above may make reference to specific hardware and software components, those skilled in the art will appreciate that different combinations of hardware and / or software components may also be used and that particular operations described as being implemented in hardware might also be implemented in software or vice versa.[0185| Computer programs incorporating various features of the present disclosure may be encoded and stored on various computer readable storage media; suitable media include magnetic disk or tape, optical storage media such as compact disk (CD) or digital versatile diskAtty. Dkt. No.: 115872-3344(DVD), flash memory, and other non-transitory media. Computer readable media encoded with the program code may be packaged with a compatible electronic device, or the program code may be provided separately from electronic devices (e.g., via Internet download or as a separately packaged computer-readable storage medium).

[0186] Thus, although the disclosure has been described with respect to specific embodiments, it will be appreciated that the disclosure is intended to cover all modifications and equivalents within the scope of the following claims.

Claims

Atty. Dkt. No.: 115872-3344WHAT IS CLAIMED IS1. A method of determining values indicating probabilities of conditions of subjects using blood samples, comprising: receiving, by one or more processors, a first plurality of images of a first blood sample having first white blood cells (WBCs) and first red blood cells (RBCs) obtained from a first subject at a first time; identifying, by the one or more processors, from the first plurality of images of the first blood sample, (i) a first set of WBC images corresponding to the first WBCs and (ii) a first set of RBC images corresponding to the first RBCs; applying, by the one or more processors, the first set of WBC images and the first set of RBC images to a machine learning (ML) architecture, wherein the ML architecture is established using a plurality of examples, each of the plurality of examples comprising (i) a second set of WBC images of a second blood sample from a second subject at a second time, (ii) a second set of RBC images of the second blood sample from the second subject at the second time, and (iii) a label indicating one of mortality or survival at a time interval relative to the second time; determining, by the one or more processors, based on applying the first plurality of images to the ML architecture, a value indicating a probability of mortality for the first subject at the time interval relative to the first time; generating, by the one or more processors, a classification of the first subject as one of one of mortality or survival in accordance with the value indicating the probability of mortality; and storing, by the one or more processors, using one or more data structures, an association between the first subject and the classification.

2. The method of claim 1, wherein generating the classification further comprises generating the classification to indicate that the first subject is to survive for the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold, and further comprising:Atty. Dkt. No.: 115872-3344 providing, by the one or more processors, an output identifying the classification to indicate that the first subject is to survive for the time interval relative to the first time.

3. The method of claim 1, wherein generating the classification further comprises generating the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold, and further comprising: providing, by the one or more processors, an output identifying the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time.

4. The method of claim 3, wherein providing the output further comprising providing, based on the classification, the output comprising at least one of: (i) a notification for a clinician to examine the first subject, (ii) a notification to administer an intervention within the time interval, or (iii) a notification for the first subject to request for medical attention.

5. The method of claim 1, further comprising receiving, by the one or more processors, a first nonimage dataset comprising at least one of (i) a first plurality of traits of the first subject, (ii) a first blood count derived from the first blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject, wherein applying to the ML architecture further comprises applying the first non-image dataset to the ML architecture, wherein at least one of the plurality of examples comprises a second non-image dataset comprising at least one of (i) a second plurality of traits of the second subject, (ii) a second blood count derived from the second blood sample, or (iii) a second plurality of measures derived from testing of the second blood sample, or (iv) a second plurality of physiological measurements of the second subject, wherein generating the classification further comprises generating the classification based on applying the first non-image dataset to the ML architecture.Atty. Dkt. No.: 115872-33446. The method of claim 1, wherein at least one of the plurality of examples further comprises the label identifying one of presence or absence of at least one of a plurality of conditions associated with the mortality in the second subject, wherein the plurality of conditions comprises a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MAHA), acidosis, multiorgan failure, anemia, hemophagocytic lymphhistiocytosis, and cytokine release syndrome, and further comprising: determining, by the one or more processors, based on applying the first plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the first subject, and wherein generating the classification further comprises generating the classification to identify one of a presence or absence of a condition of the plurality of conditions associated with the mortality in the first subject in accordance with the second value.

7. The method of claim 6, further comprising: generating, by the one or more processors, based on applying the first plurality of images to the ML architecture, a plurality of embeddings used to determine the second value; executing, by the one or more processors, using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space, each of the plurality of clusters associated with a respective set of parameters for at least one of the plurality of conditions; determining, by the one or more processors, based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters; and identifying, by the one or more processors, from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.

8. The method of claim 7, wherein at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from testing of the second blood sample,Atty. Dkt. No.: 115872-3344 wherein the clustering model is established by determining, for each cluster of the plurality of clusters, the respective set of parameters based on the plurality of parameters of the at least one example associated with a second plurality of embeddings assigned to the cluster.

9. The method of claim 1, further comprising: determining, by the one or more processors, a relative score indicating a difference between the first subject and a plurality of subjects based on the value indicating the probability of mortality for the first subject and a second value indicating a composite probability of mortality of a plurality of subjects; and identifying, by the one or more processors, from a plurality of categories, a category for the first subject in accordance with the relative score and a score category for the category.

10. The method of claim 1, further comprising providing, by the one or more processors, for presentation, a user interface comprising one or more of: (i) a category for the first subject, (ii) a relative score between the first subject and a plurality of subjects, (iii) the value indicating the probability of mortality for the first subject, (iv) a set of parameters for a condition, (v) the condition of the first subject.

11. The method of claim 1, wherein the first blood sample is acquired in accordance with peripheral blood smear (PBS), and the first plurality of images of the first blood sample is generated within a time of the PBS, wherein identifying the first set of WBC images and the first set of RBC images further comprises identifying at least one image from the first plurality of images as one of the first set of WBC images and the first set of RBC images based on a visual characteristic of the at least one image.

12. The method of claim 1, wherein the ML architecture further comprises: a WBC encoder configured to generate a set of WBC embeddings using the first set of WBC images,Atty. Dkt. No.: 115872-3344 an RBC encoder configured to generate a set of RBC embeddings using the first set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the first subject based on the set of WBC embeddings, the set of RBC embeddings, and the set of feature embeddings, and a classifier configured to generate the classification using the value.

13. The method of claim 1, wherein the time interval comprises at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 80days, 90 days, 120 days, or 180 days.

14. The method of claim 1, wherein the first subject is at risk of or diagnosed with cancer, wherein the cancer comprises at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.

15. A method of training machine learning (ML) architectures to determine values indicating expected conditions of subjects using blood samples, comprising: retrieving, by one or more processors, a training dataset including a plurality of examples, each of the plurality of examples comprising: (i) a plurality of images of a blood sample havingAtty. Dkt. No.: 115872-3344 white blood cells (WBCs) and red blood cells (RBCs) obtained from a subject at a time and (ii) a label indicating one of mortality or survival at a time interval relative to the time; identifying, by the one or more processors, from the plurality of images of the blood sample of at least one example of the plurality of examples in the training dataset, (i) a set of WBC images corresponding to the WBCs and (ii) a set of RBC images corresponding to the RBCs; applying, by the one or more processors, the set of WBC images and the set of RBC images to a machine learning (ML) architecture comprising a plurality of weights to determine a value indicating a probability of mortality for the subject at the time interval relative to the time; generating, by the one or more processors, a classification of the subject as one of mortality or survival in accordance with the value indicating the probability of mortality; comparing, by the one or more processors, the classification of the subject generated in accordance with the value determined by the ML architecture with the label of the training dataset; and updating, by the one or more processors, at least one of the plurality of weights in the ML architecture based on comparing the classification and the label.

16. The method of claim 15, wherein at least one of the plurality examples further comprises a non-image dataset including at least one of (i) a plurality of traits of the subject, (ii) a blood count derived from the blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject, wherein applying to the ML architecture further comprises applying the non-image dataset to the ML architecture, wherein generating the classification further comprises generating the classification based on applying the non-image dataset to the ML architecture.

17. The method of claim 16, wherein at least one of the plurality examples further comprises the label identifying a presence or an absence of at least one of a plurality of conditions associated with the mortality in the subject, wherein the plurality of conditions comprises a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminatedAtty. Dkt. No.: 115872-3344 intravascular coagulation (DIC), microangiopathic hemolytic anemia (MHA), acidosis, multiorgan failure, anemia, hemophagocytic lymphhistiocytosis, and cytokine release syndrome, and further comprising: determining, by the one or more processors, based on applying the plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the first subject, wherein generating the classification further comprises generating the classification to identify a presence or an absence of a condition of the plurality of conditions associated with the mortality in the subject in accordance with the second value, and wherein updating at least one of the plurality of weights further comprises updating at least one of the plurality of weights based on comparing the condition identified by the classification and the label.

18. The method of claim 17, further comprising: generating, by the one or more processors, based on applying the first plurality of images to the ML architecture, a plurality of embeddings used to determine the second value; executing, by the one or more processors, using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space, each of the plurality of clusters associated with a respective set of parameters for at least one of the plurality of conditions; determining, by the one or more processors, based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters; and identifying, by the one or more processors, from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.

19. The method of claim 18, wherein at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from a testing of a second blood sample, wherein the clustering model is established by determining, for each cluster of the plurality of clusters, a respective set of parameters based on the plurality of parameters of the at least one example associated with a second plurality of embeddings assigned to the cluster.Atty. Dkt. No.: 115872-334420. The method of claim 15, further comprising: determining, by the one or more processors, a relative score indicating a difference between the subject and a plurality of subjects based on the value indicating the probability of mortality for the subject and a second value indicating a composite probability of mortality of a plurality of subjects; and identifying, by the one or more processors, from a plurality of categories, a category for the subject in accordance with the relative score and a score category for the category.

21. The method of claim 15, wherein the blood sample is acquired in accordance with peripheral blood smear (PBS), wherein first plurality of images of the first blood sample is generated within a time of the PBS, wherein identifying the set of WBC images and the set of RBC images further comprises identifying at least one image from the plurality of images as one of the set of WBC images and the set of RBC images based on a visual characteristic of the at least one image.

22. The method of claim 15, wherein the ML architecture further comprises: a WBC encoder configured to generate a set of WBC embeddings using the set of WBC images, an RBC encoder configured to generate a set of RBC embeddings using the set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the subject based on the set of WBC embeddings, the set of RBC embeddings, the set of feature embeddings, and a classifier configured to generate the classification using the value and a non-image dataset.Atty. Dkt. No.: 115872-334423. The method of claim 15, wherein the time interval comprises at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 80days, 90 days, 120 days, or 180 days.

24. The method of claim 15, wherein the subject is at risk of or diagnosed with cancer, wherein the cancer comprises at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.

25. A system of determining values indicating probabilities of conditions of subjects using blood samples, comprising: one or more processors coupled with memory, the one or more processors are configured to: receive a first plurality of images of a first blood sample having first white blood cells (WBCs) and first red blood cells (RBCs) obtained from a first subject at a first time; identify, from the first plurality of images of the first blood sample, (i) a first set of WBC images corresponding to the first WBCs and (ii) a first set of RBC images corresponding to the first RBCs; apply the first set of WBC images and the first set of RBC images to a machine learning (ML) architecture, wherein the ML architecture is established using a plurality of examples, each of the plurality of examples comprising (i) a second set of WBC images of a second blood sample from a second subject at a second time, (ii) a second set of RBC images of the secondAtty. Dkt. No.: 115872-3344 blood sample from the second subject at the second time, and (iii) a label indicating one of mortality or survival at a time interval relative to the second time; determine based on applying the first plurality of images to the ML architecture, a value indicating a probability of mortality for the first subject at the time interval relative to the first time; generate a classification of the first subject as one of one of mortality or survival in accordance with the value indicating the probability of mortality; and store, using one or more data structures, an association between the first subject and the classification.

26. The system of claim 25, wherein the one or more processors are configured to: generate the classification to indicate that the first subj ect is to survive for the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold; and provide an output identifying the classification to indicate that the first subject is to survive for the time interval relative to the first time.

27. The system of claim 25, wherein the one or more processors are configured to: generate the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time, responsive to the value indicating the probability of mortality not satisfying a threshold; and provide an output identifying the classification to indicate that the first subject is at risk of dying within the time interval relative to the first time.

28. The system of claim 27, wherein the one or more processors are configured to provide, based on the classification, the output comprising at least one of: (i) a notification for a clinician to examine the first subject, (ii) a notification to administer an intervention within the time interval, or (iii) a notification for the first subject to request for medical attention.Atty. Dkt. No.: 115872-334429. The system of claim 25, wherein the one or more processors are configured to: receive a first non-image dataset comprising at least one of (i) a first plurality of traits of the first subject, (ii) a first blood count derived from the first blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject; apply the first non-image dataset to the ML architecture, wherein at least one of the plurality of examples comprises a second non-image dataset comprising at least one of (i) a second plurality of traits of the second subject, (ii) a second blood count derived from the second blood sample, or (iii) a second plurality of measures derived from testing of the second blood sample, or (iv) a second plurality of physiological measurements of the second subject; and generate the classification based on applying the first non-image dataset to the ML architecture.

30. The system of claim 25, wherein at least one of the plurality of examples further comprises the label identifying one of presence or absence of at least one of a plurality of conditions associated with the mortality in the second subject, wherein the plurality of conditions comprises a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MAHA), acidosis, multiorgan failure, anemia, hemophagocytic lymphhistiocytosis, and cytokine release syndrome, wherein the one or more processors are configured to: determine based on applying the first plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the first subject; and generate the classification to identify one of a presence or absence of a condition of the plurality of conditions associated with the mortality in the first subject in accordance with the second value.

31. The system of claim 30, wherein the one or more processors are configured to:Atty. Dkt. No.: 115872-3344 generate based on applying the first plurality of images to the ML architecture, a plurality of embeddings used to determine the second value; execute, using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space, each of the plurality of clusters associated with a respective set of parameters for at least one of the plurality of conditions; determine, based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters; and identify, from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.

32. The system of claim 31, wherein at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from testing of the second blood sample, wherein the clustering model is established by determining, for each cluster of the plurality of clusters, the respective set of parameters based on the plurality of parameters of the at least one example associated with a second plurality of embeddings assigned to the cluster.

33. The system of claim 25, wherein the one or more processors are configured to: determine a relative score indicating a difference between the first subject and a plurality of subjects based on the value indicating the probability of mortality for the first subject and a second value indicating a composite probability of mortality of a plurality of subjects; and identify, from a plurality of categories, a category for the first subject in accordance with the relative score and a score category for the category.

34. The system of claim 25, wherein the one or more processors are configured to provide for presentation, a user interface comprising one or more of: (i) a category for the first subject, (ii) a relative score between the first subject and a plurality of subjects, (iii) the value indicating the probability of mortality for the first subject, (iv) a set of parameters for a condition, (v) the condition of the first subject.Atty. Dkt. No.: 115872-334435. The system of claim 25, wherein the first blood sample is acquired in accordance with peripheral blood smear (PBS), and the first plurality of images of the first blood sample is generated within a predefined time of the PBS, wherein the one or more processors are configured to identify at least one image from the first plurality of images as one of the first set of WBC images and the first set of RBC images based on a visual characteristic of the at least one image.

36. The system of claim 25, wherein the ML architecture further comprises: a WBC encoder configured to generate a set of WBC embeddings using the first set of WBC images, an RBC encoder configured to generate a set of RBC embeddings using the first set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the first subject based on the set of WBC embeddings, the set of RBC embeddings, and the set of feature embeddings, and a classifier configured to generate the classification using the value.

37. The system of claim 25, wherein the time interval comprises at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 80days, 90 days, 120 days, or 180 days.

38. The system of claim 25, wherein the first subject is at risk of or diagnosed with cancer, wherein the cancer comprises at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers,Atty. Dkt. No.: 115872-3344 lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.

39. A system of training machine learning (ML) architectures to determine values indicating expected conditions of subjects using blood samples, comprising: one or more processors coupled with memory, the one or more processors configured to: retrieve a training dataset including a plurality of examples, each of the plurality of examples comprising: (i) a plurality of images of a blood sample having white blood cells (WBCs) and red blood cells (RBCs) obtained from a subject at a time and (ii) a label indicating one of mortality or survival at a time interval relative to the time; identify from the plurality of images of the blood sample of at least one example of the plurality of examples in the training dataset, (i) a set of WBC images corresponding to the WBCs and (ii) a set of RBC images corresponding to the RBCs; apply the set of WBC images and the set of RBC images to a machine learning (ML) architecture comprising a plurality of weights to determine a value indicating a probability of mortality for the subject at the time interval relative to the time; generate a classification of the subject as one of mortality or survival in accordance with the value indicating the probability of mortality; compare the classification of the subject generated in accordance with the value determined by the ML architecture with the label of the training dataset; and update at least one of the plurality of weights in the ML architecture based on comparing the classification and the label.

40. The system of claim 39, wherein at least one of the plurality examples further comprises a nonimage dataset including at least one of (i) a plurality of traits of the subject, (ii) a blood countAtty. Dkt. No.: 115872-3344 derived from the blood sample, (iii) a first plurality of parameters derived from testing of the first blood sample, or (iv) a first plurality of physiological measurements of the first subject, wherein the one or more processors are configured to: apply the non-image dataset to the ML architecture; and generate the classification based on applying the non-image dataset to the ML architecture.

41. The system of claim 39, wherein at least one of the plurality examples further comprises the label identifying a presence or an absence of at least one of a plurality of conditions associated with the mortality in the subject, wherein the plurality of conditions comprises a systemic inflammatory response syndrome (SIRS), sepsis, septic shock, bacterial infection, disseminated intravascular coagulation (DIC), microangiopathic hemolytic anemia (MHA), acidosis, multiorgan failure, anemia, hemophagocytic lymphhistiocytosis, and cytokine release syndrome, and wherein the one or more processors are configured to: determine, based on applying the plurality of images to the ML architecture, a second value indicating a probability of a condition of the plurality of conditions associated with the mortality in the subject; generate the classification to identify a presence or an absence of a condition of the plurality of conditions associated with the mortality in the subject in accordance with the second value; and update at least one of the plurality of weights based on comparing the condition identified by the classification and the label.

42. The system of claim 41, wherein the one or more processors are configured to: generate based on applying the plurality of images to the ML architecture, a plurality of embeddings used to determine the second value; execute using the plurality of embeddings, a clustering model comprising a plurality of clusters within a feature space, each of the plurality of clusters associated with a respective set of parameters for at least one of the plurality of conditions;Atty. Dkt. No.: 115872-3344 determine based on executing the clustering model, an assignment of the plurality of embeddings to a cluster of the plurality of clusters; and identify from the cluster associated with the assignment of the plurality of embeddings, a set of parameters for the condition of the plurality of conditions.

43. The system of claim 42, wherein at least one example of the plurality of examples comprises the label identifying a plurality of parameters derived from testing of a second blood sample, wherein the clustering model is established by determining, for each cluster of the plurality of clusters, the respective set of parameters based on the plurality of parameters of the at least one example associated with a second plurality of embeddings assigned to the cluster.

44. The system of claim 39, wherein the one or more processors are configured to: determine a relative score indicating a difference between the subject and a plurality of subjects based on the value indicating the probability of mortality for the subject and a second value indicating a composite probability of mortality of a plurality of subjects; and identify, from a plurality of categories, a category for the subject in accordance with the relative score and a score category for the category.

45. The system of claim 39, wherein the blood sample is acquired in accordance with peripheral blood smear (PBS), wherein first plurality of images of the first blood sample is generated within a predefined time of the PBS, wherein the one or more processors are configured to identify at least one image from the plurality of images as one of the set of WBC images and the set of RBC images based on a visual characteristic of the at least one image.

46. The system of claim 39, wherein the ML architecture further comprises: a WBC encoder configured to generate a set of WBC embeddings using the set of WBC images,Atty. Dkt. No.: 115872-3344 an RBC encoder configured to generate a set of RBC embeddings using the set of RBC images, a feature encoder configured to generate a set of feature embeddings using a non-image dataset, an aggregate predictor configured to determine the value indicating the probability of mortality for the subject based on the set of WBC embeddings and the set of RBC embeddings, and a classifier configured to generate the classification using the value and a non-image dataset.

47. The system of claim 39, wherein the time interval comprises at least one of 6 hours, 12, hours, 24 hours, 2 days, 3 days, 4 days, 5 days, 6 days, 7 days, 30 days, 45 days, 80days, 90 days, 120 days, or 180 days.

48. The system of claim 39, wherein the subject is at risk of or diagnosed with cancer, wherein the cancer comprises at least one of carcinomas, sarcomas, hematopoietic cancers, adrenal cancers, bladder cancers, blood cancers, bone cancers, brain cancers, breast cancers, carcinoma, cervical cancers, colon cancers, colorectal cancers, corpus uterine cancers, ear, nose and throat (ENT) cancers, endometrial cancers, esophageal cancers, gastrointestinal cancers, head and neck cancers, Hodgkin's disease, intestinal cancers, kidney cancers, larynx cancers, leukemias, liver cancers, lymph node cancers, lymphomas, lung cancers, melanomas, mesothelioma, myelomas, nasopharynx cancers, neuroblastomas, non- Hodgkin's lymphoma, oral cancers, ovarian cancers, pancreatic cancers, penile cancers, pharynx cancers, prostate cancers, rectal cancers, sarcoma, seminomas, skin cancers, stomach cancers, teratomas, testicular cancers, thyroid cancers, uterine cancers, vaginal cancers, vascular tumors, and metastases thereof.

Images