Computer-Implemented Method and System

Abstract

A computer-implemented method for collating patient data for analysis includes receiving a set of input data including a plurality of patient data records, the plurality of patient data records including medical image data and at least one other type of patient data; and generating a vector for each of the plurality of patient data records by processing each patient data record with a corresponding encoding algorithm, wherein the encoding algorithm used to generate the vector is selected based on the type of the patient data record, and the vector is for use by a machine learning model.

Description

[Technical field]

[0001] A computer-implemented method and system for collating patient data for analysis. [Background technology]

[0002] Assessment of a patient's health, and in particular prediction of future health, such as prediction of disease onset and disease progression, is essential to determining treatment plans. Longitudinal information (i.e., information obtained over time) in the form of a patient's health care records exists in abundance, both directly related to a specific condition and generally. With respect to general data, assessment of health or a specific condition is often influenced by the specific patient (e.g., age and sex) and lifestyle (e.g., diet and smoking status), all of which may affect the likelihood or progression of disease. If a patient is already being monitored or investigated for a condition, there are usually multiple longitudinal studies with various information and often investigation reports such as imaging data or physiological data. All of these can be used to identify risks or health problems and predict onset or progression.

[0003] However, the data is usually highly complex and heterogeneous, and making sense of this information so that it can all be factored in is difficult. For example, manually reviewing it is time-consuming, depends on the experience and expertise of a particular medical professional, and can be error-prone or inaccurate. Automated analysis is limited by the heterogeneous nature of the data.

[0004] US Patent Application Publication No. 2018144466A1 discloses a method and apparatus for improved deep learning for image acquisition. An exemplary apparatus includes a deployed learning device including a second processor for implementing a second DLN generated from a first DLN of a training learning device, the deployed learning device configured to provide second imaging system setting parameters to the imaging system in response to receiving a second input for image acquisition. Summary of the Invention

[0005] In a first aspect of the present invention, there is provided a computer implemented method of collating patient data for analysis, the computer implemented method comprising the steps of receiving a set of input data comprising a plurality of patient data records, the plurality of patient data records comprising medical image data and at least one other type of patient data; and generating a vector for each of the plurality of patient data records by processing each patient data record with a corresponding encoding algorithm, the encoding algorithm used to generate the vector being selected based on the type of patient data record, and the vector being for use by a machine learning model.

[0006]

[0005] Embodiments provide a method for pre-processing heterogeneous data types, specifically different types of patient data (patient data records), to allow subsequent processing of these data types to provide output related to the patient's health or specific condition. This allows subsequent analysis to be more accurate and sensitive since it is not limited to, and does not unintentionally weight, a particular data type. Data can be obtained from medical records (e.g., electronic medical records (EMR), radiology information systems, cardiology information systems, and / or picture archiving and communication systems (PACS)), and therefore a significant amount of heterogeneous patient data can be combined.

[0007] In particular, because the nature and location of the most important information for various medical inputs differs, using vectorization processing better preserves the important information held in each of the different data formats. For example, compared to systems that assimilate data using a single encoding algorithm, the embodiments provide a way in which the heterogeneous data from each record is encoded (i.e., a feature vector is generated) using an algorithm or model appropriate or optimized for that particular data format. This ensures that data integrity is preserved and important features of each record are maintained for subsequent processing, which can improve both the accuracy and sensitivity of subsequent analysis. Furthermore, this overcomes the problems associated with different data models, formats, and semantics associated with the data formats. The vectors may then be combined / analyzed to determine an output.

[0008] Each encoding algorithm is a model specific to a data type (e.g., specific to medical image data or a particular format of medical image data, or specific to non-medical image data or a particular format of non-image medical data). The model may be used to convert the data into a vector. The encoding algorithm is a deep learning algorithm, a traditional machine learning, signal processing, and / or a statistical model. Examples of deep learning models include, but are not limited to, convolutional neural networks (e.g., 3D convolutional networks), transformer neural networks, and fully connected networks. In some embodiments, the model is a machine learning model such as a neural network (e.g., 3D convolutional networks, fully connected networks).

[0009] This pre-processing allows machine learning models or algorithms to be used to provide outputs regarding patient health (e.g., predictions regarding the occurrence or progression of a condition), which are more accurate and more sensitive to changes due to the pre-processing steps specific to the data format.

[0010] At least one data input (i.e., patient data records or "patient data") includes medical image data, and correspondingly, generation of the vector based on the medical image data records includes using an encoding algorithm that is specific to the medical image data, and in some cases specific to the particular type of medical image data. The medical image data may include scans or frames of a particular anatomical region, including, for example, 2D, 3D, 2D+t, or 3D+t medical image data (e.g., ultrasound images, MRI images).

[0011] The at least one other type of data is a different type of data from the medical image data format (it is a different format or a different model with different semantics), such that a different encoding algorithm is used to pre-process the data. In some embodiments, the data is a different medical image data format (e.g. obtained by a different modality) and / or a non-medical image data format (e.g. as physiological data (e.g. signals such as ECG signals) or text-based medical data (e.g. consultation / diagnosis reports)). In one embodiment, the at least one other patient data format is non-medical image data related to a patient. In one embodiment of the computer-implemented method, the non-image medical data includes physiological data. In one embodiment of the computer-implemented method, the non-medical image data is in the form of at least one of text-based data, signal data, and tabular data.

[0012] The data received and analyzed relates to a single patient, and the output will therefore relate to a single patient. The data used in the methods may be from a single time point (e.g., a single study or survey) or may be from multiple time points (e.g., different studies or surveys occurring at different times (e.g., multiple days)). The data used may relate to a particular condition (e.g., studies directed to monitoring or assessing a condition) or may have consideration of more general health conditions that contribute to a particular condition.

[0013] In one embodiment, the computer-implemented method further comprises analysing the vector using a machine learning model and generating an output comprising information relating to the patient's health. Thus, in such an embodiment, the method is for providing or generating an output comprising information relating to the patient's health.

[0014] The use of machine learning through a machine learning engine or model (which may be a classification / regression model using, for example, a traditional machine learning model or a deep learning model) provides additional benefits since the output can be more accurate and sensitive as more subtle changes over time can be sensed. In embodiments, the machine learning model receives input in the form of feature vectors and uses the machine learning engine to determine the output, for example changes over time between data (e.g., images). The machine learning model is trained such that the machine learning model is capable of providing the output. In some embodiments, a training database is provided that contains training information including images and associated outputs.

[0015] In one embodiment, the method further comprises combining the vectors prior to analyzing the vectors. Combining the output vectors may be performed using a number of different techniques, including stacking the vectors as one-dimensional or multidimensional vectors. Thus, in one embodiment of the computer-implemented method, combining the vectors comprises combining the vectors to form one-dimensional or multidimensional vectors. The ability of the computer-implemented method in some embodiments to process multiple types of data over time, including both image, signal, and non-image data, allows for a significant improvement in the accuracy of previous attempts to predict disease onset or progression.

[0016] In one embodiment, the output of the computer-implemented method includes a prediction regarding the diagnosis, onset, and / or progression of a disease in a patient. Producing an output that includes information about the diagnosis, onset, and / or progression of a disease is beneficial because the output may be used by a physician to obtain information that would not otherwise be readily available. Having a predictive model that considers vectors that include multiple patient data records rather than a single record allows for more accurate predictions to be made regarding the diagnosis, onset, and / or progression of a disease. Thus, in such an embodiment, the method is for providing or generating a prediction regarding the diagnosis, onset, and / or progression of a disease in a patient.

[0017] In one embodiment of the computer-implemented method, the machine learning model is a recurrent and / or regression model, such as a recurrent neural network. An embodiment in which the data used in the method is acquired at different time points (i.e., longitudinal data) and a regression / recurrent model (such as a recurrent neural network) is used may be particularly advantageous, since temporal features may be taken into account. For example, the time at which the data was acquired may be used to weight the analysis. Examples include gated recurrent units (GRUs), long short-term memory (LSTM), or other regression models.

[0018] In one embodiment, the plurality of patient data records includes patient data records acquired at multiple time points, and the step of analyzing the vector using the machine learning model includes weighting each patient data record based on the time point at which the record was created. In some embodiments, the weighting gives more weight to more recently acquired data. An example of this is weighting a scan taken more heavily than a scan taken at an earlier time point. This allows for improved accuracy of prediction regarding diagnosis, onset, and / or progression of disease. Although some risk scores for disease prediction (e.g., Framingham risk score) have emerged, these are general and rely on only one study or survey. An embodiment that considers longitudinal data for disease prediction provides improved accuracy over these risk scores by considering data acquired over time and by monitoring changes.

[0019] This is also particularly advantageous since there is often a large amount of longitudinal information in patient medical records, and health care systems more generally, that can be used for such analyses.

[0020] In an embodiment where the patient data includes data acquired at multiple time points (e.g., different dates or studies), each patient data includes a timestamp associated therewith, and the generation of the vector includes a time-resolved feature vector based on the timestamps. In such an embodiment, subsequent processing determines a disease onset / progression prediction based on the multiple time-resolved feature vectors. In these embodiments, the vectors may be analyzed using regression-based models, such as recurrent / recurrent neural networks (RNN) and / or long short-term memory (LSTM), which may be used to analyze the temporal information to provide more accurate and improved output.

[0021] In one embodiment, the machine learning model is trained. Training may occur prior to or as part of analysis of the vectors using the machine learning model. In one embodiment, training is performed using medical data duplets, each medical data duplets including (i) multiple patient data records for a respective patient, and (ii) disease diagnosis and / or progression data for that patient.

[0022] The computer-implemented method can process several forms of data, including medical image data, and therefore there is a large pool of potential training data obtained over time that can be used to train this machine learning model.

[0023] In one embodiment, generating vectors for each of the plurality of patient data records by processing each patient data record with a corresponding encoding algorithm is performed using neural network models, traditional machine learning models, signal processing, and / or statistical models. These models, such as neural networks, are trained with data such that performance improves over time. Improving the encoding algorithm has the potential to reduce the processing power and time required to generate the vectors. In one embodiment, the neural network is a convolutional neural network, a transformer neural network, or a fully connected neural network. Convolutional neural networks are particularly effective for processing medical image data, ensuring that important information is preserved in the vectors.

[0024] In one embodiment, the medical image data is in the form of two-dimensional or three-dimensional data.

[0025] In one embodiment, the plurality of patient data records includes patient data records acquired at multiple time points, and generating a vector for each of the plurality of patient data records comprises generating a time-resolved feature vector.

[0026] In one embodiment, the patient data record includes patient demographic data, which may be scalar data resulting in a vector of scalar data.

[0027] In one embodiment, the plurality of patient data records includes (i) a first patient data record including at least one medical image (i.e., medical image data), (ii) a second patient data record including non-image medical data including physiological signals, and (iii) a third patient data record including text-based medical data. This combination of data is traditionally particularly difficult to combine, but the disclosed method provides improved pre-processing using these three types of data to enable more accurate determination of health parameters.

[0028] In a second aspect of the present invention, there is provided a system for collating patient data for analysis, the system comprising: a memory comprising instruction data representing a set of instructions; and one or more processors in communication with the memory and configured to execute the set of instructions, the set of instructions, when executed by the processor, causing the processor to perform the computer-implemented method described above.

[0029] When referring to the processing performed in the above methods, it should be understood that the methods are not limited to being performed by a single processor or computer. Instead, the methods may be performed by multiple processors located at different locations and interconnected, for example, using a distributed network. An example of this is the processing of images from multiple data sources transferred or stored in a distributed network (e.g., a "cloud" based network). Furthermore, any processing steps may occur in different processors within the network.

[0030] The processor is implemented in any suitable manner, together with software and / or hardware, to perform the various required functions. Examples of processor components employed in various embodiments of the present disclosure include, but are not limited to, conventional microprocessors, application specific integrated circuits (ASICs), and field programmable gate arrays (FPGAs). The processor is associated with one or more non-transitory storage media, such as volatile and non-volatile computer memories, such as RAM, PROM, EPROM, and EEPROM. The non-transitory storage media is encoded with one or more programs that perform the required functions when executed in the one or more processors and / or controllers. The various storage media are fixed or portable within the processor or controller, such that the one or more programs stored therein can be loaded into the signal processing unit, the characterization unit, and / or the processor.

[0031] In an embodiment, the system comprises a user interface, such as a display for displaying the output. Alternatively or additionally, the system comprises a communication interface device, such as a wireless transmitter, configured to transmit the output to an external device, such as a personal computer, tablet, smartphone, remote server, etc.

[0032] In a third aspect, there is provided a method of collating patient data for analysis (such as a method of predicting the diagnosis, progression or onset of a disease), the method comprising the steps of: (i) obtaining input data comprising a plurality of patient data records, the plurality of patient data records comprising medical image data and at least one other type of patient data, and (ii) generating a vector for each of the plurality of patient data records by processing each patient data record using a corresponding encoding algorithm, the encoding algorithm used to generate the vector being selected based on the type of the patient data record, and the vector being for use by a machine learning model, using a computer-implemented method as described with respect to any embodiment defined herein.

[0033] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiment(s) described hereinafter.

[0034] For a better understanding of the present invention, and to show more clearly how the same may be carried into effect, reference will now be made, by way of example only, to the accompanying drawings in which: [Brief description of the drawings]

[0035] [Figure 1] 2 is a schematic diagram of a computer-implemented method according to one embodiment of the present invention. [Diagram 2] FIG. 4 is a schematic diagram of a computer-implemented method according to another embodiment of the present invention. [Diagram 3] FIG. 4 is a schematic diagram of a computer-implemented method according to another embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0036] It should be understood that the detailed description and specific examples, while indicating exemplary embodiments of the devices, systems, and methods, are intended for purposes of illustration only and are not intended to limit the scope of the invention. These and other features, aspects, and advantages of the devices, systems, and methods of the present invention will become better understood from the following description, the appended claims, and the accompanying drawings. It should be understood that the figures are merely schematic and not drawn to scale. It should also be understood that the same reference numerals are used throughout the figures to denote the same or similar parts.

[0037] Variations of the disclosed embodiments may be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps, and singular elements do not exclude a plurality.

[0038] A single processor or other unit may fulfill the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0039] It should be noted that when the term "adapted" is used in the claims or the specification, it is intended to be equivalent to the term "configured."

[0040] Any reference signs in the claims should not be construed as limiting the scope.

[0041] FIG. 1 provides a general overview of a computer-implemented method 100 for collating patient data for analysis, which is used to provide an output 140 that includes information related to the patient's health.

[0042] As shown, the method 100 has two separate branches corresponding to two different patient data types (pertaining to a single patient). Specifically, the method 100 includes receiving a set of input data including receiving medical image data 110 and receiving non-medical image patient data 110'. In an embodiment, the medical image data is in the form of 2D, 3D, 2D+t, or 3D+t data (such as ultrasound or MRI image data). In some embodiments, the non-medical data may be text-based data (e.g., text-based medical test data, clinical findings, diagnoses, and patient demographic information) or physiological signals.

[0043] Thus, the data received is heterogeneous. In this embodiment, the medical image data and the non-medical image data are related in that they have relevance to the health condition under investigation. The input data is extracted or received from a health care database or multiple databases. For example, the data is extracted or received from EMRs, radiology information systems, cardiology information systems, and / or PACS. Alternatively, it is extracted or received from devices or equipment used to acquire the data, such as medical devices or sensors. Interfacing with clinical, operational, and demographic data sources can be done by any available prior art IT communication protocol, such as HL7, DICOM, and FHIR.

[0044] Each data type is prepared (pre-processed) separately. Thus, the method 100 further comprises generating a first vector 120 for the medical image data 110 received in the first step. The step of generating the first vector 120 comprises processing each patient data record (i.e., the medical image data in this case) using an encoding algorithm that is optimized and specific for medical image data. The encoding algorithm is an AI or machine learning algorithm or model. Examples of those for medical image data include, but are not limited to, convolutional neural networks (e.g., 3D convolutional networks) and fully connected networks.

[0045] The method 100 also comprises generating a second vector 120' for the non-medical image data 110' received in the first step. In this case, generating the vector 120' comprises processing each patient data record (i.e., the non-medical image data in this case) using an encoding algorithm that is optimized and specific to a particular data type, that is a different encoding algorithm than that used to encode the first vector, such that the data extraction is optimized for each data type, thereby providing an output that is particularly accurate and in which important features and information of each particular data type are preserved and taken into account in subsequent processing.

[0046] Thus, for each type of patient data, a model appropriate to the data type can be constructed. These models convert data from different sources into vectors.

[0047] The method 100 further comprises analysing the vectors using a machine learning model 130. This comprises analysing and combining the vectors to provide an output 140 that includes information relating to the patient's health that takes into account the different inputs received in the first steps 110, 110'.

[0048] The output may be in multiple forms. It may include a health indicator (e.g., a diagnosis) or may include a prediction (e.g., expected progression) regarding the diagnosis, onset, and / or progression of a disease. It may include medical images or models showing particular regions of interest or the like. The output may be provided to a display of an associated system.

[0049] The general method 100 of Figure 1 may be used to process different types of data acquired at a single time point (i.e., as part of a single study or study of a patient), or may be used to process different types of data acquired at different times (e.g., acquired as part of different studies or studies of a patient, or across a period of time). In other embodiments, the data is acquired at different times, but without regard to the time it was acquired or recorded. In such embodiments, the data is treated as if it were acquired at a single time point.

[0050] If the data is acquired at a single time point (or if time information is not considered), the output vectors from the step of generating vectors 120, 120' may be fed into a regression / classification model for training. This may further include using historical disease diagnosis and / or progression information, and the classification / regression model may be trained using traditional machine learning models or deep learning models.

[0051] If data is acquired at multiple time points and time factors (i.e., temporal features) are to be considered, the output vector from the step of generating vectors 120, 120' may be fed to a recurrent / regressive model (e.g., a recurrent neural network, RNN, and / or long short-term memory (LSTM)) for disease training. The combining step may first include combining multiple time vectors as a stacked one-dimensional or multidimensional vector. The use of regressive models such as RNN and / or long short-term memory (LSTM) may be used to analyze the temporal information to provide a more accurate and improved output. The vector generated in step 130 may be a time-resolved feature vector.

[0052] 2 illustrates a second embodiment of a method 200 for providing a prediction regarding the occurrence of a disease or condition. In this embodiment, there are multiple data types and within each data type there are multiple patient records taken at different times. This is shown as two main branches corresponding to the two different data types and two sub-branches, each corresponding to the different times at which the data was taken.

[0053] As shown in the first branch, the method 200 comprises receiving a set of input data comprising receiving 210 medical image data acquired at a first time point, and receiving 210A medical image data of the same type acquired at a second, different time point, for example a first image corresponding to a scan of an anatomical region acquired on a first date, and a second image corresponding to the same type of scan of the same region acquired on a later date, as is common in monitoring the progression of a disease.

[0054] The input data also includes non-medical image patient data, such that the method also has a step 210' of receiving non-medical image patient data acquired at a first time point (which does not have to correspond to the same time point when the medical image data was acquired), and a step 210A' of receiving non-medical image patient data acquired at a second time point (which does not have to correspond to the same time point when the medical image data was acquired).

[0055] The method 200 further comprises the step of generating 220, 220', 200A' vectors based on the input data. Specifically, each type of data is prepared (processed) separately.

[0056] In this embodiment, the method 200 further comprises a step 220 of generating vectors for the medical image data received in the first step 210, 210A. This step comprises processing each patient data record (i.e., in this case, the medical image data) using an encoding algorithm that is optimized and specific to the medical image data. In this embodiment, the date the medical image data was acquired is encoded into the composite vector to produce a time-resolved vector. In one embodiment, and by way of example only, these vectors may be combined into a single time-resolved vector prior to analysis as part of step 220. In other embodiments, these vectors may be provided as separate vectors to the analytical model in the next step.

[0057] The method also includes a step 220', 220A' of generating a second vector for the non-medical image data received in the first step 210', 210A'. This step includes processing each patient data record (i.e., multiple non-image data in this case) using an encoding algorithm that is optimized and specific to the particular format of non-image data. In this embodiment, the date the data was acquired is encoded into the composite vector to produce a corresponding time-resolved vector. In this embodiment, and by way of example only, these vectors are provided as separate vectors to the analytical model in the next step. In other embodiments, these vectors may be combined into a single time-resolved vector prior to analysis.

[0058] The method 200 further comprises analyzing the vectors using a machine learning model 230. This includes analyzing and combining the vectors to provide an output 240 including information related to the patient's health that takes into account the different inputs and temporal characteristics of the data received in the first step 210, 210A, 210', 210A'.

[0059] The machine learning model used in the analysis step of the above embodiment may be trained using pairs of information. Each medical data pair may include multiple data sets of different types of medical data for each patient and disease diagnosis and / or progression data for each of said patients. This allows the model to be trained to determine indicators of a particular disease and improve the accuracy of prediction. If temporal data is used, this may also be used to refine predictions regarding the occurrence or progression of the condition.

[0060] FIG. 3 illustrates generally a further embodiment of a method 300 for providing a prediction regarding the occurrence of a patient condition.

[0061] The method in this embodiment receives a set of input data 310, 310', 310", which includes three separate sets of data corresponding to three different time points. Thus, the method comprises receiving first medical data 310 relating to a patient acquired at a first time point, receiving second medical data 310' relating to a patient acquired at a second time point, and receiving third medical data 310" relating to a patient acquired at this time point.

[0062] In this embodiment, the first medical data includes medical image data 311, medical data related to physiological signals 312, and text-based medical data related to diagnostic reports 313, all acquired at a first time point as part of a first study. The second medical data includes medical image data 311', and text-based medical data related to diagnostic reports 313', all acquired at a second time point as part of a second study. The third medical data includes medical image data 311" and medical data related to physiological signals 312" all acquired at a third time point as part of a third study.

[0063] In this embodiment, method 300 processes each data type within each study (study by study) separately using a predefined and data type specific encoding algorithm, thereby generating a separate vector for each data type within each study. The multiple vectors for each study are then combined (within each study) to provide a multi-dimensional or stacked vector for each study. These are each encoded with time specific information to allow for temporal analysis.

[0064] Specifically, the method 300 further comprises generating vectors for a first data set 320, which itself comprises generating a first vector for medical image data 321 using a medical image coding model (i.e. a model adapted or configured to convert medical image data into vectors), generating a second vector for medical data relating to physiological signals 322 using a physiological signal coding model (i.e. a model adapted or configured to convert physiological signal data into vectors), and generating a third vector for text-based medical data 323 using a text-based data coding model (i.e. a model adapted or configured to convert text-based data into vectors). These are then combined 325 to provide a multi-dimensional or stacked vector encoding the temporal information. Using a model specific to the data format ensures that all important information present in each of the data is preserved and encoded into the vectors.

[0065] The method 300 also includes generating vectors for a second set 320' comprising generating a first vector for the medical image data 321' using the medical image coding model and generating a second vector for the text-based medical data 323' using the text-based data coding model, which are then combined 325' to provide a multi-dimensional or stacked vector that encodes the temporal information.

[0066] The method 300 further comprises generating vectors for a third data set 320″ comprising generating a first vector for the medical image data 321″ using the medical image coding model and generating a second vector for medical data relating to physiological signals 322″ using the physiological signal coding model, which are then combined 325″ to provide a multi-dimensional or stacked vector encoding temporal information.

[0067] The resultant vectors from this process are then combined and analyzed in the next processing step 330. In particular, the vectors are used to train a machine learning model, which in this embodiment comprises a recurrent neural network that analyzes and takes into account the temporal characteristics of the data. The method in this embodiment includes weighting each patient data record based on the time at which the record was created.

[0068] Results from the analysis are output 340 in the form of a prediction regarding the occurrence of the patient condition, taking into account the three studies and the data within each study.

[0069] Although the invention has been described with reference to specific embodiments and examples above, it will be understood that modifications can be made to the embodiments and examples without departing from the invention.

Claims

1. 1. A computer-implemented method for collating patient data for analysis, the computer-implemented method comprising: receiving a set of input data including a plurality of patient data records, the plurality of patient data records including medical image data and at least one other type of patient data; generating a vector for each of the plurality of patient data records by processing each patient data record with a corresponding encoding algorithm, wherein the encoding algorithm used to generate the vector is selected based on the type of the patient data record and is a model specific to the data type of the patient data record, and wherein the vector is for use by a machine learning model; 10. A computer-implemented method comprising:

2. analyzing the vector using the machine learning model; generating an output including information regarding the health of a patient; The computer-implemented method of claim 1 further comprising:

3. The computer-implemented method of claim 2 , further comprising the step of combining the vectors prior to the step of analyzing the vectors.

4. 4. The computer-implemented method of claim 3, wherein combining the vectors comprises combining the vectors to form a one-dimensional or multi-dimensional vector.

5. The computer-implemented method of claim 2 , wherein the output is a prediction regarding the diagnosis, onset, and / or progression of a disease in a patient.

6. The computer-implemented method of claim 2 , wherein the machine learning model is a recursive and / or regression model.

7. the plurality of patient data records includes the patient data records obtained at multiple time points; 3. The computer-implemented method of claim 2, wherein analyzing the vectors using the machine learning model comprises weighting each patient data record based on when the patient data record was created.

8. The machine learning model is trained using medical data pairs, each medical data pair comprising: said plurality of patient data records for each patient; said patient's disease diagnosis and / or progression data; The computer-implemented method of claim 2 , comprising:

9. The computer-implemented method of claim 1 , wherein the at least one other type of patient data is non-medical image data relating to a patient.

10. The computer-implemented method of claim 9 , wherein the non-image medical data comprises physiological data.

11. The computer-implemented method of claim 9 , wherein the non-medical image data is in the form of at least one of text-based data, signal data, and tabular data.

12. 2. The computer-implemented method of claim 1, wherein generating the vector for each of the plurality of patient data records by processing each patient data record with the corresponding encoding algorithm is performed using a neural network model, a conventional machine learning model, signal processing, and / or a statistical model.

13. 13. The computer-implemented method of claim 12, wherein the neural network is a convolutional neural network, a transformer neural network, or a fully connected neural network.

14. the plurality of patient data records includes the patient data records obtained at multiple time points; The computer-implemented method of claim 1 , wherein generating the vector for each of the plurality of patient data records comprises generating a time-resolved feature vector.

15. 1. A system for collating patient data for analysis, the system comprising: a memory containing instruction data representing a set of instructions; and one or more processors in communication with the memory and configured to execute the set of instructions, the set of instructions, when executed by the processors, causing the processors to perform the computer-implemented method of any one of claims 1 to 14.

Images