An artificial intelligence (AI) based system and method for pre-operative assessment and surgical outcome prediction

Abstract

The present disclosure proposes an artificial intelligence (AI) based System (100) for preoperative assessment and surgical outcome prediction. The AI-based system (100) comprises one or more client modules collect data from electronic medical records (EMRs), Data repositories and Databases via a data collection module (122), then standardize and pre-process it with a standardization & pre-processing module (124) for integrity and compatibility. The pre-processed data is transmitted to a server (104) through a network(102) using an API module (126) and thereafter forwarded to an input module (112). The input module (112) authenticates and validates data of at least one patient. The data is subsequently sent to a processing module (114), and a prediction module (116) for predicting surgical duration and blood loss, and the extreme Gradient Boost model for postoperative patient placement predictions.

Description

4. DESCRIPTION:Field of the invention:

[0001] The present disclosure generally relates to the technical field of healthcare and, in specific, relates to an artificial intelligence (Al)-based system and method of preoperative assessment and estimation of crucial surgical outcomes like surgical duration, blood loss, and postoperative patient placement.Background of the invention:

[0002] Surgery is a medical procedure that uses specialized techniques to access a body and treat a variety of conditions. Surgery is a crucial tool in modern medicine, allowing doctors to diagnose and treat patients. Surgeries are complex procedures typically performed in a sterile operating room by a team of healthcare professionals.

[0003] Efficient management of operating rooms (ORs) and accurate pre-operative assessments are pivotal in optimizing surgical outcomes and resource utilization in healthcare organizations. The high costs associated with surgeries and OR operations make them a significant focus for hospital administrators seeking to enhance efficiency and reduce expenses. Effective OR management requires balancing resources to prevent underutilization, which can lead to staff idle time and increased patient waiting, and overutilization, which can cause staff overload, patient dissatisfaction, and increased risk of errors. A crucial component in this balancing act is the accurate prediction of surgical duration and other related outcomes.

[0004] Surgeries involve various procedures supported by a team that includes surgeons, anesthesiologist's, nurses, and other staff. These surgeries can be classified as either emergency or elective, and their duration significantly impacts OR scheduling and management. Traditional practices often rely on physicians to predict surgery duration based on their experience, which can lead to inaccuracies and suboptimal scheduling. Alternatively, some hospitals use default Duration of Surgery times calculated as the mean duration for specific surgery types, which also fails to capture the variability in individual cases.2

[0005] In recent years, many studies around the world have reported the use of various machine-learning (ML) methods to accurately predict surgery duration. Linear regression techniques have been explored using patient and surgical factors and have reported the importance of such variables in predicting the total surgical procedure time. Distributional modelling methods such as Kernel Density Estimation (KDE) as well as log-normal distributions were also demonstrated to be able to effectively predict surgery duration. More complex methods such as heteroscedastic neural network regression combined with expressive drop-out regularized neural networks have also been shown to have good performance.

[0006] Existing surgical outcome prediction technologies often demonstrate suboptimal accuracy, potentially leading to unnecessary delays or postponements in surgical procedures. This directly impacts patient health and disrupts the efficient flow of hospital operations. Prevailing pre-anesthesia evaluations typically focus on a restricted set of risk factors, hindering their ability to comprehensively assess a patient's surgical risk profile. Unpredictable surgical outcomes often hinder the optimal allocation of resources, leading to operational challenges and increased costs for healthcare institutions.

[0007] Most current systems lack an integrated feedback mechanism from healthcare professionals, hindering the continuous refinement and improvement of predictive models. Existing models may exhibit potential biases due to insufficient inclusivity and fairness, limiting their generalizability and effectiveness across diverse patient populations. The high financial burden associated with established pre-Anesthesia assessment tools and operating room scheduling systems can significantly limit their adoption, particularly in resource- constrained settings. Deficiencies in preoperative assessment can lead to delays or postponements of surgical procedures. This not only has adverse consequences for the patient's physical and emotional well-being but also disrupts the established surgical schedule and hinders the overall efficiency of hospital operations.

[0008] Therefore, there is a need for an Al-based system and method for pre-Anesthesia assessment where machine learning and data analysis methods are utilised to predict surgical outcomes such as surgery duration, surgery blood loss, and postoperative patient placement. There is also a need for a system that incorporates a broader scope of riskfactors, leading to a more thorough and accurate evaluation of a patient's comprehensive preoperative assessment. There is also a need for a system that empowers clinicians to allocate resources efficiently, minimizing operational challenges and reducing associated costs for healthcare institutions. There is also a need for a system that minimizes the risk of surgery delays or postponements.

[0009] Addressing these needs, Apollo Hospitals Group has developed an advanced Al- based Preoperative Assessment & Surgical outcome prediction Tool. The tool is developed using a comprehensive dataset, which was retrospectively collected from over 3,23,000 patients at pre-anesthesia checks. However, 1,17,900 patient data was excluded due to missing fields, leaving a sample size of 2,05,100 patients. These patients underwent surgeries across eight Apollo group centers.

[0010] The dataset comprises anonymized demographic data and over 40 input features Including but not limited to vital signs, comorbidities, Airway & Anesthesia assessments, lab results, and surgical details. Out of these, more than 30 predictors were deemed significant based on feature importance and correlation coefficient. They are demographic details & vitals include age, gender, oxygen saturation, respiratory rate (RR), and blood pressure (BP), body mass index (BMI), comorbidities & clinical state cover lung disease, arrhythmia, hypertension, coronary artery disease (CAD), transient ischemic attack (TIA), cerebrovascular accident (CVA), seizures, heart failure, diabetes, Interstitial Lung Illness (I LI), central nervous system (CNS) conditions, immunocompromised status. Laboratory parameters covers hemoglobin levels, international normalized ratio (INR), lymphocyte count, platelet count, and white blood cell (WBC) count. Surgical details encompass anesthesia type, surgical site. Anesthesia Assessment Include airway intervention, chest imaging, American society of anaesthesiologist's physical status (ASA PS) patient classification, metabolic equivalent of task (MET) category and obstructive sleep apnea (OSA) score & pre-operative ward status.

[0011] Harnessing Linear and Bayesian Neural Network regression models, the tool predicts surgical duration and blood loss. It also employs the extreme Gradient Boost model for postoperative patient placement predictions. The tool achieved an adjusted R2of 0.92 [Cl 0.909 - 0.939] for predicting surgical duration, an adjusted R2of 0.66 [Cl 0.649 - 0.671] forpredicting operative blood loss, and an AUC of 0.83 [Cl 0.819 - 0.841] for predicting postoperative ward placement.

[0012] This tool integrates a continuous feedback loop from patient outcomes, facilitating ongoing refinement. Key predictive factors include Body Mass Index (BMI) [5.54 (Cl 5.42 - 5.65)], International Normalized Ratio (INR) [4.74 (Cl 4.65 - 4.84)], and Heart Failure status [1.41 (Cl 1.32 - 1.50)], among others. These factors are integral to accurately estimating longer surgical durations (>20% higher than the upper limit).

[0013] The preoperative assessment and surgical outcome prediction tool use deep learning algorithms to provide reliable forecasts for surgeons, anaesthesiologists, and healthcare teams. The tool's integration into the preoperative planning process improves scheduling accuracy and resource allocation efficiency. This advancement in preoperative assessment has a significant impact on patient outcomes and the efficiency of surgical operations. Using large datasets, the tool is intended to improve risk assessment, resulting in increased efficiency and effectiveness in surgical healthcare delivery.Objectives of the invention:

[0014] The invention aims to provide an Al-based system and method for preoperative assessment, using advanced machine learning methods to predict critical surgical outcomes like surgery duration, blood loss, and post-operative placement during pre-anesthesia checks

[0015] The other objective of the invention is to provide a system that provides improved accuracy in surgical outcome prediction to reduce surgery delays and postponements, thereby improving patient care and streamlining hospital operations.

[0016] The other objective of the invention is to provide a system that incorporates a broader scope of risk factors that leads to a more comprehensive and precise assessment of a patient's profile to predict the surgical outcomes.

[0017] Another objective of the invention is to provide a system that empowers clinicians to allocate resources efficiently, minimizing operational challenges and reducing associated costs for healthcare institutions and thereby promoting patient satisfaction.

[0018] The invention aims to develop a system that can accurately interpret a wide range of surgical pr and patient groups, ensuring high predictive accuracy even when new data is added. This broadens the system's application across a variety of hospital settings and patient populations.Summary of the invention:

[0019] The present disclosure proposes an artificial intelligence (Al) based system and method for pre-operative assessment and surgical outcome prediction during preanesthesia checks. The following presents a simplified summary in order to provide a basic understanding of some aspects of the claimed subject matter. This summary is not an extensive overview. It is not intended to identify key / critical elements or to delineate the scope of the claimed subject matter. Its sole purpose is to present some concepts in a simplified form as a prelude to the more detailed description that is presented later.

[0020] In order to overcome the above deficiencies of the prior art, the present disclosure is to solve the technical problem to provide an artificial intelligence (Al)-based system for preoperative Assessment and predicting the estimates for surgical duration, blood loss, and postoperative patient placement there by promoting resource allocation.

[0021] According to one aspect, the invention provides an artificial intelligence (Al)-based configured to predict key surgical Outcomes at pre-anesthesia checks. The system supports integration with a consumer's computing device through various approaches, including the use of REST APIs. This integration method ensures seamless operation and generally employs HTTPS for secure communication. The artificial intelligence (Al) based preoperative assessment system can be accessed on a computing device connected to a server via a network. The server, housing both processor and memory components, is linked to a database for efficient data management. Within the system, a plurality of modules is executed by the processor. These modules include an input module, a processing module, a prediction module, a clinical pathway module, and a risk prediction response module.

[0022] In one embodiment, one or more client modules are configured to collect input data of a patient from various sources such as, but not limited to, electronic medical records (EMRs), Data repositories and Databases through a data collection module. Subsequently,the input data is standardized and pre-processed by a standardization and preprocessing module to ensure integrity and compatibility to obtain pre-processed data. Thereafter, the pre-processed data is transmitted to the server via the network using an API integration module. The server transmits the pre-processed data to the input module.

[0023] In one embodiment, the input module is configured to initiate authentication and validation of the pre-processed data, ensuring compliance with required formats. The input module is configured to enable users to facilitate the input data including, but not limited to, personal details such as age, gender, height, weight, BMI, and vital signs like heart rate, temperature, respiratory rate, and blood pressure. The input module further facilitates input of current medical history such as lung diseases, arrhythmia, heart failure, central nervous system disorders, coronary artery disease, diabetes mellitus, hypertension, interstitial lung disease, symptoms and immunocompromised states.

[0024] Additionally, the input module incorporates anesthesia assessment data from the ASA physical status classification system, pre-operative ward situation, and oxygen saturation, possible need for airway intervention, chest imaging, obstructive sleep Apnea score, and metabolic equivalent of task score. It also records COVID-19 vaccination history. Lab parameters like haemoglobin levels, platelet count, INR, lymphocyte count, white blood cell count, random blood sugar and HbAlc, AST (Aspartate Transaminase), ALT (Alanine Transaminase), ALP (Alkaline Phosphatase) levels, sodium and potassium levels, and creatinine levels are also included. Surgical procedure details like surgical specialty, surgery name, and Anesthesia technique are also recorded. Thedata from input Module is transmitted to processing Module for processing.

[0025] In one embodiment, the processing module transforms and standardizes data from the input module to meet the necessary input criteria for the subsequent analysis. In one embodiment, the prediction module is structured to provide Key surgical outcomes by leveraging regression models such as linear and Bayesian neural networks for predicting surgical duration and blood loss, and the extreme Gradient Boost model for postoperative patient placement predictions. In one embodiment, the clinical pathway module is designed to provide a recommended protocol of subsequent actions for at least one patient, based on the estimates of surgical outcomes. In one embodiment, the risk predictionresponse module presents estimates of surgery duration & blood loss along with contributing risk factors plot, postoperative patient placement.

[0026] According to another aspect, the invention provides a method for comprehensive pre-anesthesia assessment and predicting key surgical outcomes. At one step, the client modules are configured to collect the input data from various sources through a data collection module, with the data subsequently being standardized and pre-processed by a standardization and preprocessing module to ensure integrity and compatibility. Thereafter, the pre-processed data is transmitted to the server via a network using API integration module. The server transmits the pre-processed data to the input module. The input module initiates the authentication and validation of the pre-processed data, facilitating the inclusion of personal details, vital signs data, current medical history, and anesthesia assessment data, laboratory parameters, and surgical procedure details. At another step, the processing module standardizes and transforms the data from the input module to meet the necessary criteria for the subsequent analysis.

[0027] Further, at another step, the prediction module uses regression models such as linear and Bayesian neural networks for predicting surgical duration and blood loss, and the extreme gradient boost model for postoperative patient placement predictions. In another step, the clinical pathway module generates a recommended protocol of next actions for a patient, based on the estimates of surgical outcomes. At further step, the risk prediction response module presents estimates of surgery duration, blood loss along with contributing risk factors plot, and postoperative patient placement.

[0028] Further, objects and advantages of the present invention will be apparent from a study of the following portion of the specification, the claims, and the attached drawings.Detailed description of drawings:

[0029] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate an embodiment of the invention, and, together with the description, explain the principles of the invention.

[0030] FIG. 1 illustrates a block diagram of an artificial intelligence (Al)-based system for preoperative assessment and predicting surgical outcomes, in accordance to an exemplary embodiment of the invention.

[0031] FIG. 2 illustrates a flowchart of a method for preoperative assessment and predicting surgical outcomes using the Al-based system, in accordance to an exemplary embodiment of the invention.

[0032] FIGs. 3A-3C illustrate user interface screenshots for predicting surgical outcomes through the Al-based system, in accordance to an exemplary embodiment of the invention.

[0033] FIG. 4 forest plot illustrating odds ratios (ORs) and 95% confidence intervals (Cis) for factors affecting surgery duration, showing ORs and their impact on duration, in accordance to an exemplary embodiment of the invention.

[0034] FIG. 5 forest plot depicting ORs and Cis for factors affecting the prediction of blood loss, showing significant factors and their impact, in accordance to an exemplary embodiment of the invention.

[0035] FIG. 6A illustrates performance evaluation of different models for predicting surgery duration, including metrics such as mean square error (MSE), R2, and adjusted R2, in accordance to an exemplary embodiment of the invention.

[0036] FIG. 6B illustrates bias-variance trade-off surface plot capturing the relationship between bias and variance in predicting surgery duration, visualizing model performance adjustments, in accordance to an exemplary embodiment of the invention.

[0037] FIG. 6C illustrates training and validation loss curves for the surgery duration prediction model, showing performance improvement over epochs, in accordance to an exemplary embodiment of the invention.

[0038] FIG. 6D illustrates scatter plot of predicted vs. actual surgery durations, assessing model accuracy and alignment with real-world outcomes, in accordance to an exemplary embodiment of the invention.

[0039] FIG. 7A illustrates performance metrics comparison for different models predicting blood loss, including mean squared error (MSE) and R2 values, in accordance to an exemplary embodiment of the invention.

[0040] FIG. 7B illustrates bias-variance trade-off surface plot capturing the relationship between bias and variance in predicting blood loss prediction, visualizing model performance adjustments, in accordance to an exemplary embodiment of the invention.

[0041] FIG. 7C illustrates training and validation loss curves for the blood loss prediction model, indicating model learning and generalization, in accordance to an exemplary embodiment of the invention.

[0042] FIG. 7D illustrates scatter plot of predicted vs. actual blood loss, showing model performance and accuracy in predicting blood loss during surgery, in accordance to an exemplary embodiment of the invention.

[0043] FIG. 8A illustrates performance metrics of the XGB model for post-operative patient placement, including accuracy, AUC, sensitivity, and specificity, in accordance to an exemplary embodiment of the invention.

[0044] FIG. 8B illustrates feature importance analysis for predicting post-operative patient placement, highlighting the impact of various features, in accordance to an exemplary embodiment of the invention.

[0045] FIG. 8C illustrates ROC curve for the XGB model's performance in post-operative patient placement, showing model discrimination ability, in accordance to an exemplary embodiment of the invention.

[0046] FIG. 8D illustrates precision-recall curve for the XGB model in post-operative patient placement, illustrating the trade-off between precision and recall, in accordance to an exemplary embodiment of the invention.

[0047] FIG. 9 illustrates system architecture for the Al-based pre-anesthesia assessment tool, detailing the development and deployment process, in accordance to an exemplary embodiment of the invention.Detailed invention disclosure:

[0048] Various embodiments of the present invention will be described in reference to the accompanying drawings. Wherever possible, same or similar reference numerals are used in the drawings and the description to refer to the same or like parts or steps.

[0049] The present disclosure has been made with a view towards solving the problem with the prior art described above, and it is an object of the present invention to provide an Al- based system that predicts estimates for surgical duration, blood loss, and postoperative patient placement.

[0050] According to an exemplary embodiment of the invention, FIG. 1 refers to a block diagram of an artificial intelligence (Al)-based system 100 for predicting surgical outcomes. In one embodiment, the Al-based system 100 supports integration with a consumer's computing device through various approaches, one of the methods is use of REST APIs. This integration method ensures seamless operation and generally employs HTTPS for secure communication. The Al-based system 100 for preoperative assessment and predicting estimates of key surgical outcomes. The Al-based system 100 can be accessed on a computing device connected to a server 104 via a network 102. The server 104, housing both processor 106 and memory 108 components, is linked to a database 110 for efficient data management. Within the Al-based system 100, a plurality of modules is executed by the processor 106. These modules include an input module 112, a processing module 114, a prediction module 116, a clinical pathway module 118, and a risk prediction response module 120.

[0051] In one embodiment, one or more client modules are configured to collect input data of one or more patients from various sources such as, but not limited to, electronic medical records (EMRs), Data repositories and Databasesthrough a data collection module 122. The collected data is subsequently standardized and pre-processed by a standardization and pre-processing module 124 to ensure data integrity and compatibility. Thereafter, the pre- processed data is transmitted to the server 104 via the network 102 utilizing an API integration module 126. The server 104 then transmits the pre-processed data to the input module 112.11

[0052] In one embodiment, the input module 112 is configured to initiate authentication and validation of the pre-processed data, ensuring compliance with required formats, the input module 112 is further configured to enables the medical practitioners or the patient to provide personal details of the patientincluding, but not limited to, age, gender, height, weight, BMI, and vital signs data like heart rate, temperature, respiratory rate, and blood pressure. The input module 112 is further configured to receivethe input data related to current medical history of the patient including, but not limited to, lung diseases, arrhythmia, heart failure, central nervous system disorders, coronary artery disease, diabetes mellitus, hypertension, and interstitial lung disease. The module also facilitates the entry of data on immunocompromised states, and cardiovascular conditions such as chest pain, irregular heartbeat, blood clots. It further accommodates data on respiratory disorders like asthma, emphysema, and sleep apnea, along with thyroid and gastrointestinal issues like liver diseases and reflux. Additionally, the module allows for the input of data on blood disorders, including bleeding tendencies, sickle cell disease, and any objections to blood transfusions.

[0053] Furthermore, the input module 112 further incorporates anesthesia assessment data from the ASA physical status classification system, pre-operative ward situation, and oxygen saturation, possible need for airway intervention, chest imaging, obstructive sleep apnea score, and metabolic equivalent of task (MET) score. The input module 112 further records COVID-19 vaccination history. Lab parameters like haemoglobin levels, platelet count, INR, lymphocyte count, white blood cell count, random blood sugar and HbAlc, AST (Aspartate Transaminase), ALT (Alanine Transaminase), ALP (Alkaline Phosphatase) levels, sodium and potassium levels, and creatinine levels and pregnancy test are also included. Surgical procedure details like surgical specialty, surgery name, and anesthesia technique are also considered. Data collected by the input module 112 is transmitted to the processing module 114 for processing.

[0054] In one embodiment, the processing module 114 is configured to modify and standardize the pre-processed data from the input module 112 to meet the necessary input criteria for the subsequent analysis. In one embodiment, the prediction module 116 is adapted to provide key surgical outcomes by leveraging regression models such as linear and Bayesian neural networks for predicting surgical duration and blood loss, and theextreme Gradient Boost model for postoperative patient placement predictions. In one embodiment, the clinical pathway module 118 is adapted to provide a recommended protocol of subsequent actions for at least one patient, based on the predicted surgical outcomes, like the need for transfusions of blood and blood products, in accordance with the anticipated blood loss. In one embodiment, the risk prediction response module 120 is configured to provide estimates of surgery duration & blood loss along with contributing risk factors plot, postoperative patient placement.

[0055] In another exemplary embodiment of the invention, as illustrated in FIG. 2, a flowchart 200 represents the method for predicting key surgical outcomes using the Al- based system 100. Initially, at step 202, the client modules are configured to collectthe input data of a patient from various sources such as, but not limited to, electronic medical records (EMRs), Data repositories and Databases through a data collection module 122. At step 204, the input data is subsequently standardized and pre-processed by the standardization and preprocessing module 124 to ensure integrity and compatibility, thereby obtainpre- processed data. Subsequently, at step 206, the pre-processed data is transmitted to the server 104 via the network 102 through the API integration module 126. The server 104 then transmits the pre-processed data to the input module 112 for further processing. At step 208, the input module 112 enables authentication and validation ofthe pre-processed data related to one or more personal details, vital signs data, current medical history, anesthesia assessment data, laboratory parameters, and surgical procedure details. Subsequently, the input module 112 providesvalidated data of the patient.

[0056] Subsequently, at step 210, the processing module 114 standardizes and modifies the validated data to ensure it meets the necessary criteria for the subsequent analysis. At step 212, the prediction module 116 employs regression models such as linear and Bayesian neural networks to predict surgical duration and blood loss, and uses the extreme gradient boost model for postoperative patient placement predictions. Then, at step 214, the clinical pathway module 118 creates a Recommended Protocol of future actions for a patient, based on the estimated surgical outcomes. Finally, at step 216, the risk prediction response module 120 delivers estimates of the surgery duration, blood lossestimates, contributing risk factors, and postoperative patient placement, and generating a PDF report containing these predictions.

[0057] In another exemplary embodiment of the invention, FIGs. 3A to 3C depict sample user interface screenshots (300, 302, and 304) for forecasting surgical outcomes via the Al- based system 100. FIG. 3A represents a user interface diagram where a healthcare professional, after logging into the Al-based system 100 web interface, can enter the patient's data, including personal details. Additional input parameters can be entered as shown in FIG. 3B. In one embodiment, the patient data input includes personal variables and vital signs such as age, gender, height, weight, BMI, heart rate, temperature, respiratory rate, and blood pressure. It also collects data on current medical history, encompassing lung diseases, arrhythmia, heart failure, central nervous system disorders, coronary artery disease, diabetes mellitus, hypertension, and more. It further accommodates information on cardiovascular conditions, respiratory disorders, thyroid and gastrointestinal issues, and blood disorders.

[0058] Additionally, the input module 112 integrates anesthesia assessment data, preoperative ward situation, oxygen saturation, possible need for airway intervention, chest imaging, sleep apnea score, and MET score. It records COVID-19 vaccination history, lab parameters including hemoglobin levels, platelet count, INR, lymphocyte count, white blood cell count, random blood sugar, HbAlc, AST, ALT, ALP levels, sodium and potassium levels, creatinine levels, and pregnancy status. Information about the surgical procedure and Anesthesia technique is also considered. The prediction module 116 processes the input patient's data to provide an output of surgery duration, surgery blood loss, post-operative ward Placement, and clinical decision support system. In specific, the prediction module analyses more than 30 comprehensive input parameters and utilizes regression models such as Linear and Bayesian Neural Networks for predicting surgical duration and blood loss, and the extreme Gradient Boost model for postoperative patient placement predictions.

[0059] The estimates of key surgical outcome predictions generated by the prediction module 116 is provided to the healthcare professional through the user interface, as shown in FIG. 3C. The user is provided with estimates of key surgical outcomes such as estimated surgery duration & blood loss during surgery along with contributing risk factors graph and post-operative ward placement. Further, a printed report or PDF copy can be generated.

[0060] Numerous advantages of the present disclosure may be apparent from the discussion above. In general, surgical interventions are complex and multifaceted, involving various variables that can influence the duration of the operation and the amount of blood loss. In accordance with the present disclosure, the Al-based system 100 is disclosed that aids the users to do holistic preoperative assessment and estimate key surgical outcomes such surgical duration & blood loss during surgery along with contributing risk factors graph, and postoperative patient placement. The Al-based system 100 is designed to optimize patient outcomes, resource utilization, and overall efficiency in surgical procedures.

[0061] In another exemplary embodiment of the invention, as illustrated in FIG. 4, a forest plot 400 displays the odds ratios (ORs) along with the 95% confidence intervals (Cis) for various factors affecting surgery duration. Each factor is listed in the left column, with corresponding event and total case counts in both experimental and control groups. The OR (odds ratio) indicates the likelihood of an event in the experimental group relative to the control group, while the 95% Cl provides the range within which can be 95% confident the true OR lies. Squares represent each factor's OR, with the size reflecting the study's weight or estimate precision, and the horizontal line through the square denoting the 95% Cl. The vertical line at OR=1 signifies the point of no effect. ORs greater than 1 suggest increased odds, and those less than 1 suggest decreased odds.

[0062] Heart failure and coronary artery disease (CAD) both show an OR of 1.41, with heart failure at 1.41 [1.32, 1.50] and CAD at 1.41 [1.38, 1.43], suggesting 41% increased odds for each condition. Diabetes, with an OR of 1.23 [1.21, 1.26], and obstructive sleep apnea (OSA), with an OR of 1.20 [1.18, 1.22], indicate 23% and 20% increased odds of extended surgery duration, respectively. The site of surgery has an OR of 1.25 [1.22, 1.27], while age shows an OR of 1.26 [1.23, 1.28], indicating 25% and 26% increased odds, respectively. Body mass index (BMI) presents a significantly high OR of 5.54 [5.42, 5.65], and saturation has an OR of 1.36 [1.28, 1.44], indicating 36% increased odds. The international normalized ratio (INR) also shows a high OR of 4.74 [4.65, 4.84]. The common effect model reveals an overall OR of 1.28 [1.28, 1.29] for all factors combined, suggesting a general increase in odds across the studied factors. The random effects model, accounting for variability between studies, shows an OR of 1.19 [1.01, 1.40]. Heterogeneity statistics indicate substantial variabilityamong the included studies, with an I2of 100%, T2of 0.1942, and a statistically significant p- value of 0.

[0063] In another exemplary embodiment of the invention, as illustrated in FIG. 5, a forest plot 500 illustrates the odds ratios (ORs) and 95% confidence intervals (Cis) for various factors affecting the prediction of blood loss. Each factor is listed in the left column, with corresponding event and total case counts in both experimental and control groups. The OR indicates the likelihood of an event in the experimental group relative to the control group, while the 95% Cl provides the range within which can be 95% confident the true OR lies. Squares represent each factor's OR, with the size reflecting the study's weight or estimate precision, and the horizontal line through the square denoting the 95% Cl. The vertical line at OR=1 signifies the point of no effect; ORs greater than 1 suggest increased odds, and those less than 1 suggest decreased odds.

[0064] For the prediction of blood loss, several factors show significant odds ratios. Heart failure and coronary artery disease (CAD) both demonstrate an OR of 2.50, with heart failure at 2.69 [2.42, 2.99] and CAD at 2.50 [2.35, 2.65]. Diabetes presents an OR of 1.98 [1.85, 2.12]. Obstructive sleep apnea (OSA) shows an OR of 1.28 [1.20, 1.35]. Age has an OR of 1.35 [1.27, 1.43]. Body mass index (BMI) has a significantly high OR of 1.21 [1.13, 1.30], indicating a substantial increase in the likelihood of blood loss. Oxygen Saturation shows an OR of 0.98 [0.81, 1.18]. The international normalized ratio (INR) also shows a high OR of 1.03 [0.89, 1.20], suggesting a significant impact on blood loss prediction. The common effect model reveals an overall OR of 1.22 [1.20, 1.24] for all factors combined, suggesting a general increase in odds across the studied factors. The random effects model, accounting for variability between studies, shows an OR of 1.07 [0.85, 1.36]. Heterogeneity statistics indicate substantial variability among the included studies, with an I2of 100%, T2of 0.4223, and a statistically significant p-value of 0.

[0065] In another exemplary embodiment of the invention, as illustrated in FIG. 6A, the performance evaluation of three distinct models 600 for predicting surgery duration is demonstrated. The models assessed include the linear neural network, Bayesian neural network, and best set performance. The evaluation metrics provide a comprehensive view of their effectiveness. Mean square error (MSE) quantifies the accuracy of predictions, witha lower MSE indicating better performance in minimizing prediction errors. R2 (Coefficient of Determination) measures how well the model explains the variance in the dependent variable, with values closer to 1 implying a stronger fit of the model to the data. Adjusted R2, similar to R2, accounts for the number of predictors in the model, offering a more accurate assessment of model performance by considering potential overfitting.

[0066] The data sets used for evaluation include a test set comprising 205,100 data points with 33 features, and a Validation Set including 38,027 data points with the same 33 features. Specifically, the linear neural network demonstrates reasonable accuracy and a good fit to the test data, with an MSE of 1111.99, an R2 of 0.877, and an adjusted R2 of 0.876. The Bayesian neural network performs even better, with a lower MSE of 669.83 and higher R2 and adjusted R2 values of 0.926. Unfortunately, the specific MSE value for the best set performance is not visible due to image cut off, warranting further investigation. For the validation set, similar limitations due to the image dimensions exist, but the available metrics still provide valuable insights into model performance.

[0067] In another exemplary embodiment of the invention, as illustrated in FIG. 6B, the bias-variance trade-off surface plot 602 visually captures the delicate balance between bias and variance in predicting surgery duration. This three-dimensional surface plot provides a comprehensive view of how variations in bias and variance affect overall model performance. Bias refers to the error introduced by approximating a real-world problem with a simplified model, which can lead to under fitting, while variance refers to the model's sensitivity to fluctuations in the training data, potentially causing overfitting. Striking the right balance between these factors is crucial for developing effective predictive models for clinical applications, ensuring reliable and accurate predictions.

[0068] In another exemplary embodiment of the invention, as illustrated in FIG. 6C, the training and validation loss during the model's training process 604. The blue line represents the training loss, showing how the model's performance improves over epochs. As training progresses, the loss generally decreases, indicating a better fit to the training data. The orange line reflects the model's performance on unseen validation data, and although it exhibits fluctuations, the overall trend is downward, suggesting that the model generalizes well beyond the training set. The x-axis represents the number of training epochs, rangingfrom 0 to 100, and the y-axis corresponds to the loss value. The training loss starts at around 700 and gradually decreases to just above 600. Similarly, the validation loss begins at about 750 and decreases more dramatically, showing significant fluctuations but ultimately trending downwards to below 500.

[0069] From these observations, several implications arise. First, the model's learning dynamics are evident, as the reduction in training loss indicates that the model is effectively learning from the training data and improving its ability to predict surgery durations. The downward trend in validation loss suggests that the model is not overfitting, as it performs reasonably well on unseen data. This balance between training and validation loss indicates a good balance between overfitting and under fitting.

[0070] If the training loss continued to decrease while the validation loss increased, it would signal overfitting, whereas if both losses remained high, it would indicate under fitting. Practitioners often monitor these loss curves during training, employing early stopping based on validation loss to prevent overfitting and ensure the choice of the final model considers both training and validation performance. The key takeaway from the FIG. 6C is that the model shows promising signs of effective learning and generalization. The consistent downward trend in both training and validation losses suggests that the model is improving its predictive capabilities without overfitting to the training data. This provides confidence in the model's potential utility for accurately predicting surgery durations in a clinical setting.

[0071] In another exemplary embodiment of the invention, as illustrated in FIG. 6D visualizes the relationship between predicted and actual surgery durations 606. In the scatter plot each green dot represents a surgery instance, with the x-axis representing the ground truth (actual) surgery duration and the y-axis representing the model's predicted surgery duration. The red dashed line represents a perfect prediction line where the predicted durations match the actual durations exactly. Ideally, the green dots should align along this diagonal line, indicating accurate predictions. The closer the points lie to this line, the better the model's performance. This scatter plot allows us to assess how well the model's predictions match real-world outcomes, providing a crucial tool for evaluating the model's practical applicability in clinical settings.

[0072] Several implications can be drawn from FIG. 6D. The accuracy of predictions is indicated by the proximity of the green dots to the red diagonal line, with dots close to the line implying accurate predictions where the model's estimates align closely with actual surgery durations. The concentration of dots along the diagonal suggests that the model captures the underlying patterns well, demonstrating robustness. Outliers or significant deviations from the diagonal may indicate cases where the model struggles, such as with extreme surgery durations. Practically, this scatter plot informs clinicians about the model's reliability. Consistent alignment of predictions with actual outcomes can aid in surgical planning and resource allocation, providing a valuable tool for clinical decision-making.

[0073] From the scatter plot, it is evident that while many predictions are close to the actual values, some deviations exist, especially at higher duration values. This indicates that the model performs reasonably well for most cases but may require further refinement to handle extreme values more accurately. This visualization is essential for understanding the model's strengths and limitations in a clinical context.

[0074] In another exemplary embodiment of the invention, as illustrated in FIG. 7A, the performance evaluation of various models 700 for predicting blood loss during surgery, evaluating the linear model, linear neural network, and Bayesian neural network. The metrics considered are mean squared error (MSE), R2 (coefficient of determination), and adjusted R2. MSE measures prediction accuracy, with lower values indicating better performance. R2 values closer to 1 imply a better fit of the model to the data, while adjusted R2 provides a more accurate assessment by accounting for the number of predictors. The data sets used for evaluation include a test set comprising 205,100 data points with 33 features and a validation set comprising 38,027 data points with the same 33 features. The linear neural network shows an MSE of 3728.95 and an R2 of 0.577, whereas the Bayesian neural network performs better with an MSE of 2098.84 and an R2 of 0.669. These metrics indicate that the Bayesian neural network aligns more closely with actual blood loss during surgery, suggesting its superior predictive capability.

[0075] In another exemplary embodiment of the invention, as illustrated in FIG. 7B presents a visual representation of blood loss prediction based on two independent variables (bias and weight) and one dependent variable (Mean Square Error, MSE). The three-dimensionalsurface plot 702 illustrates how variations in bias and weight impact the MSE, with a color gradient representing the predicted values across different conditions or parameter values.

[0076] The x-axis represents bias, ranging from -10 to 10. The y-axis represents weight, ranging from 0 to 20. The z-axis represents the MSE, ranging from 0 to 3 million. The surface plot uses a color gradient to indicate the MSE values, with lighter colors representing lower MSE values and darker colors representing higher MSE values. The plot demonstrates the impact of varying bias and weight on the MSE, providing a visual representation of how to optimize these parameters for more accurate blood loss predictions. Identifying regions with minimal MSE can help clinicians adjust their approaches to achieve better predictive accuracy. This visualization aids in understanding the balance between bias and variance, crucial for developing robust predictive models in clinical settings.

[0077] In another exemplary embodiment of the invention, as illustrated in FIG. 7C, the training and validation loss 704 during the training process of the blood loss prediction model. The blue line represents the training loss, indicating how the model's performance improves over epochs. As training progresses, the loss decreases, showing that the model is learning and fitting better to the training data. The orange line represents the validation loss, reflecting the model's performance on unseen validation data. Although there are fluctuations, the overall trend is downward, suggesting that the model generalizes well beyond the training set. The x-axis represents the number of training epochs, ranging from 0 to 100, and the y-axis corresponds to the loss value, which starts at approximately 22,000 and decreases to around 12,000.

[0078] Several implications arise from these observations. The reduction in training loss indicates that the model is effectively learning from the training data and improving its ability to predict blood loss. The downward trend in validation loss suggests that the model is not overfitting, as it performs reasonably well on unseen data. This balance between training and validation loss indicates a good balance between overfitting and underfitting. If the training loss continued to decrease while the validation loss increased, it would signal overfitting, whereas if both losses remained high, it would indicate underfitting. Practitioners often monitor these loss curves during training, employing early stopping based on validation loss to prevent overfitting and ensure the choice of the final modelconsiders both training and validation performance. The key takeaway from the FIG. 7C is that the model shows promising signs of effective learning and generalization. The consistent downward trend in both training and validation losses suggests that the model is improving its predictive capabilities without overfitting to the training data. This provides confidence in the model's potential utility for accurately predicting blood loss in a clinical setting.

[0079] In another exemplary embodiment of the invention, as illustrated in FIG. 7D visualizes the relationship between predicted and actual blood loss 706 during surgery. Each green dot represents a surgery instance, with the x-axis representing the ground truth (actual) blood loss and the y-axis representing the model's predicted blood loss. The red dashed line represents a perfect prediction line where the predicted blood loss matches the actual blood loss exactly. Ideally, the green dots should align along this diagonal line, indicating accurate predictions. The closer the points lie to this line, the better the model's performance. This scatter plot allows us to assess how well the model's predictions match real-world outcomes, providing a crucial tool for evaluating the model's practical applicability in clinical settings. In another exemplary embodiment of the invention, as illustrated in FIG. 8A, the performance metrics 800 of a predictive model for post-operative patient placement using an XGB (Extreme Gradient Boosting) Model are examined. The metrics evaluated include Accuracy, AUC (Area under the Curve), Sensitivity, and Specificity.

[0080] Accuracy measures the overall correctness of predictions, with a value of 0.72 for the model. This indicates that 72% of the predictions made by the model are correct. AUC (Area under the Curve) represents the model's ability to discriminate between positive and negative outcomes, with a value of 0.73. This metric is crucial as it combines sensitivity and specificity into a single performance measurement, providing insight into the model’s capability to distinguish between different classes. Sensitivity, also known as the true positive rate, indicates the proportion of actual positive cases correctly predicted by the model. In this case, the model has a sensitivity value of 0.71, meaning it accurately identifies 71% of the true positive cases. Specificity, or the true negative rate, reflects the proportion of actual negative cases correctly predicted by the model. The model shows a specificity value of 0.78, suggesting it correctly identifies 78% of the true negative cases.

[0081] An evaluation data sets consist of a test set with 205,100 data points and 33 features, and a validation set with 38,027 data points and the same 33 features. The results from the test set show an accuracy of 0.72, an AUC of 0.73, a sensitivity of 0.71, and a specificity of 0.78. These metrics indicate that the model performs well in predicting postoperative patient placement, showing a balanced ability to correctly identify both positive and negative cases. The model demonstrates reasonable accuracy in predicting postoperative patient placement, with the AUC value indicating effective discrimination between positive and negative outcomes. The sensitivity and specificity values reflect the model's balanced ability to correctly identify both positive and negative cases, further supporting its reliability in a clinical setting. These performance metrics suggest that the model is useful for assisting clinicians in making informed decisions regarding patient placement after surgery, ultimately contributing to better patient management and outcomes.

[0082] In another exemplary embodiment of the invention, as illustrated in FIG. 8B provides an analysis of feature importance 802 in determining post-operative patient placement, with features ranked by their F score to reflect their impact on the model's predictions. The type of pre-operative ward, such as ICU or general ward, holds the highest importance, significantly influencing post-operative care decisions. Platelet count is also crucial due to its impact on clotting and bleeding risks, which clinicians consider when deciding patient placement. The average duration of standard procedures plays a role, as longer procedures can affect post-operative care requirements.

[0083] Hemoglobin, measuring blood oxygen-carrying capacity, is an important feature in the predictions. Lymphocyte count, reflecting immune system health, contributes to the model's accuracy. White blood cell count (WBC) is relevant for managing infection risks and overall immune function. Body mass index (BMI) offers insights into overall health and nutritional status, which can impact post-operative recovery. Respiratory rate affects oxygenation and recovery, influencing placement decisions. Patient age is a factor in determining care needs, with different requirements for younger and older patients. Lastly, the International Normalized Ratio (INR), which measures blood clotting ability, is relevant for assessing surgical risk. This analysis highlights the top features and their significance in predicting post-operative patient placement.

[0084] In another exemplary embodiment of the invention, as illustrated in FIG. 8C presents the Receiver Operating Characteristic (ROC) curve 804, which assesses the performance of the XGB model in predicting post-operative ward placement. The ROC curve plots the True Positive Rate (TPR), also known as sensitivity or recall, against the False Positive Rate (FPR), which is calculated as 1 minus specificity. This visualization illustrates the model's ability to correctly identify positive cases while minimizing false positives.

[0085] The Area under the Curve (AUC) is a critical metric derived from the ROC curve, representing the model's overall performance. The AUC value of 0.73 indicates a reasonable level of discrimination by the model between positive and negative cases. A higher AUC suggests better model performance, with 0.73 reflecting effective capability in distinguishing between different outcomes in post-operative ward placement scenarios.

[0086] In another exemplary embodiment of the invention, as illustrated in FIG. 8D illustrates the Precision-Recall Curve (PRC) 806, which evaluates the trade-off between precision and recall (sensitivity) across various threshold values for the XGB model predicting post-operative patient placement. The PRC graphically represents how precision, the proportion of true positive predictions among all positive predictions, and recall, the proportion of actual positive cases correctly predicted by the model, change with different decision thresholds. The curve highlights the best Fl score threshold, where precision and recall are balanced, resulting in the highest Fl score, which is the harmonic mean of both metrics.

[0087] The analysis identifies additional specific thresholds: the best precision and recall thresholds are both 0.244, while the optimal Fl score threshold is 0.282. These thresholds are crucial for clinicians to determine the most suitable balance between precision and recall based on their priorities. Understanding these thresholds helps clinicians make informed decisions regarding patient placement by optimizing the trade-off between accurately identifying positive cases and minimizing false positives. Overall, the PRC and its associated thresholds offer valuable insights into balancing performance metrics effectively for post-operative patient placement.

[0088] In accordance with another exemplary embodiment of the present invention, FIG. 9 provides a schematic depiction of the system architecture 900 for a method devised todevelop and deploy the Al-based pre-anesthesia Assessment Tool. The diagram highlights a sequence of operations integral to the system's development and deployment. In step 902, the data sourcing phase gathers information from diverse sources such as, but not limited to, pre-Anesthesia checks, surgical details, and lab data from Electronic Medical Records (EMR) servers 914. Additional sources comprise clinical knowledge bases 916 such as Standard Operating Protocols (SOPs), literature from research papers and journals, and the opinions and assistance of clinicians 918. APIs are used to facilitate seamless integration and retrieval of data from these various sources.

[0089] In step 904, the data ingestion phase involves collecting data and integrating it into a centralized database 920. The data goes through a data pipeline 922 and is stored in a data repository 924 in structured formats, crucial for organizing the data and preparing it for further pre-processing stages. In Step 906, the data pre-processing phase includes metadata management 926, ETL processes 928, and data transformation and harmonization 930. These processes enhance data quality and consistency, preparing the data for subsequent analysis and model training.

[0090] In step 908, the data analysis phase, a range of statistical tools are harnessed to extract insights and correlations from the gathered data. Propensity matching 932 is used to expose hidden patterns, while descriptive statistics 934 summarize the essential characteristics of the data. The relationships between various risk factors are established through correlation coefficients 936, and the comparison of the likelihood of different outcomes is facilitated by odds / hazard ratios 938. Time-to-event data is visually represented through KM plots and survival charts 940, and complex data representations are simplified using visualization tools 942. Collectively, these techniques serve to enhance feature selection and the development of the model.

[0091] In step 910, the model development phase, the Al-based pre-anesthesia Assessment Tool employs several sophisticated models for precise predictions about key surgical outcomes such as surgery duration, blood loss, and postoperative patient placement. For predicting the ' Duration of the Surgery and the Estimated Blood Loss during the Surgery, the tool utilizes regression models, specifically the Linear Neural Network and the Bayesian Neural Network. These networks comprise multiple layers with varying numbers of neurons.The first layer consists of 128 neurons, employing a 'relu' activation function and L2 regularization to prevent overfitting. This is followed by a dropout layer with a rate of 0.1, which further aids in preventing overfitting. The subsequent layers consist of 64 and 32 neurons respectively, each with the 'relu' activation function and L2 regularization. Another dropout layer with a rate of 0.1 follows the third layer.

[0092] The final layers in these models are linear layers with one neuron, used for regression output, enabling accurate estimation of the duration of the procedure and the amount of blood loss. In addition to these neural network models, the tool also employs the extreme Gradient Boost (XGB) classification model for predicting postoperative patient placement. After evaluating various models, these were concluded to be optimal for the given predictions.

[0093] In step 912, the deployment phase integrates the model for usage through REST API 952 protocols. This architecture utilizes an API management service 944 and an application service resource 950 for model inference. The model inference code is developed in Python programming language 954. All resources are securely hosted in a Virtual Private Network 948. The API service 946 acts as the interface for the REST API, facilitating communication between the API Management Service and the web application. The API Management Service includes storage capabilities 958 for securely managing and storing model artifacts, data inputs, and results. In one such application of the integration, the model inference API is integrated into a web application code 956, deployed as a Web Application 960. This application provides a user-friendly platform for clinicians to input data and obtain preanesthesia Assessment & Key Surgical Outcome predictions, making the model operational and accessible.

[0094] The Al-based system 100 operates as an adjunct tool for physicians or healthcare professionals by leveraging Deep learning algorithms on comprehensive datasets, including patient history, surgical procedure details, and other pertinent factors. The Al-based system 100 provides a robust and reliable tool for surgeons, anaesthesiologists, and healthcare teams. The integration of predictive models can assist in preoperative planning, resource allocation, and preoperative assessment, ultimately contributing to improved patient outcomes and operational efficiency in the surgical setting. The Al-based system lOOalignswith the broader goal of advancing healthcare technologies, enhancing decision-making processes, and promoting a data-driven approach to surgical management.

[0095] The Al-based system 100 achieved an adjusted R2of 0.92 for predicting surgical duration, an adjusted R2of 0.66 for predicting operative blood loss, and an AUG of 0.83 for predicting postoperative ward placement. Based on predicted surgical outcomes, the clinical pathway module 118 provides personalized protocols, and the risk prediction response module 120 provides surgical outcome predictions such as expected surgery duration, estimated blood loss during surgery along with contributing risk factors graph, and postoperative ward placement. The Al-based system 100 and method for pre-operative assessment and surgical outcome prediction operates as a complementary tool for clinicians, enhancing informed decision-making without replacing clinicians or diagnostic tests. It adheres to ISO 13485 standards, ensuring Patient Safety and reliability. As a software as a medical device (SaMD) certified by ISO 13485, the Al-based system 100 integrates algorithms for early intervention, thereby improving cost-effective utilization of surgical staff and operating rooms and decrease patients' waiting time.

[0096] The Al-based system 100 is developed utilizing a robust deep learning algorithms. Leveraging a comprehensive dataset encompassing numerous clinical and laboratory parameters, the Al-based system 100 demonstrates potential for optimizing resource allocation, particularly in cost-constrained environments. The Al-based system 100 exhibits high accuracy in predicting critical surgical outcomes such as surgical duration, blood loss, and postoperative patient disposition.

[0097] It will readily be apparent that numerous modifications and alterations can be made to the processes described in the foregoing examples without departing from the principles underlying the invention, and all such modifications and alterations are intended to be embraced by this application.26

Claims

5. CLAIMS: l / We Claim:

1. An artificial intelligence (Al) based system (100) for preoperative assessment and predicting surgical outcomes during pre-anesthesia checks, comprising: a server (104) in communication with a database (110) for efficient data management and a user's computing device through a network (102), wherein the server (104) comprises a processor (106) and a memory (108) coupled to the processor (106), wherein the memory (108) comprises instructions and plurality of modules executable by the processor (106), wherein the plurality of modules comprises: one or more client modulesconfigured to collect input data of a patient from various sources such as electronic medical records (EMRs), Data repositories and Databases through a data collection module 122; a standardization and pre-processing module (124) configured to standardized and pre-process the collected input data to ensure integrity and compatibilitybased on system's processing requirements, thereby obtain pre-processed data; an API integration module (126) configured to transmit the pre-processed data from the standardization and pre-processing module (124) to the server (104) via the network (102); an input module (112) configured to receive the pre-processed data through the server (104) via the network (102) to enable authentication and validation of the pre-processed data, thereby ensuring compliance with required formats from the various sources, wherein the input module (112) is configured to allow one or more users to facilitate data related to one or more personal details, vital signs data, current medical history, anesthesia assessment data, laboratory parameters, and surgical procedure details, thereby obtaining validated data of the patient; a processing module (114) configured to standardize and modify the validated datato attain necessary criteria for a subsequent analysis;27a prediction module (116) configured to provide key surgical outcomes by leveraging regression models such as linear and Bayesian neural networks for predicting surgical duration and blood loss, and the extreme gradient boost model for postoperative patient placement predictions; a clinical pathway module (118) configured to provide a recommended protocol of subsequent actions for at least one patient, based on the predicted surgical outcomes, like the need for transfusions of blood and blood products, in accordance with the anticipated blood loss and thereof; and a risk prediction response module (120) configured to present predicted estimates of surgery duration & blood loss along with contributing risk factors plot, postoperative patient placement and a PDF of report can be generated, thereby acting as a complementary tool for clinicians, wherein the risk prediction response module (120) enhances resource allocation and supports clinicians without replacing clinicians or diagnostic tests, whereby, the Al-based system (100) adheres to ISO 13485 standards, ensuring patient safety and reliability, functioning as a software as a medical device (SaMD) and certified by ISO 13485, thereby the Al based pre-Anesthesia system (100) provides a robust and reliable tool for clinicians, the integration of predictive models assist in preoperative planning, resource allocation.

2. The Al-based system (100) as claimed in claim 1, wherein the input module (112) receives patient data include personal details such as age, gender, height, weight, BMI, and vital signs data heart rate, temperature, respiratory rate, and blood pressure.

3. The Al-based system (100) as claimed in claim 2, wherein the patient data comprises current medical history include lung diseases, arrhythmia, heart failure, central nervous system disorders, coronary artery disease, diabetes mellitus, hypertension, and interstitial lung disease, wherein the input module (112) facilitates entry of data on immunocompromised states, and cardiovascular conditions such as chest pain, irregular heartbeat, blood clots, data on respiratory disorders like asthma, emphysema, and sleep apnea, along with thyroid and gastrointestinal issues like liver diseases and reflux, anddata on blood disorders, include bleeding tendencies, sickle cell disease, and objections to blood transfusions.

4. The Al-based system (100) as claimed in claim 2, wherein the patient data include anesthesia assessment data such as ASA physical status classification system, preoperative ward situation, oxygen saturation, possible need for airway intervention, chest imaging, obstructive sleep apnea score, metabolic equivalent of task score, and records COVID-19 vaccination history.

5. The Al-based system (100) as claimed in claim 2, wherein the patient data include lab parameters such as hemoglobin levels, platelet count, international normalized ratio (INR), lymphocyte count, white blood cell count, random blood sugar and hemoglobin A1C (HbAlc), AST (Aspartate Transaminase), ALT (Alanine Transaminase), ALP (Alkaline Phosphatase) levels, sodium and potassium levels, and creatinine levels, pregnancy test, surgical procedure details like surgical specialty, surgery name, and anesthesia technique.

6. The Al-based system (100) as claimed in claim 1, wherein the Al-based system (100) achieved an adjusted R2of 0.92 for predicting surgical duration, an adjusted R2of 0.66 for predicting operative blood loss, and an AUC of 0.83 for predicting postoperative ward placement.

7. The Al-based system (100) as claimed in claim 1, wherein the surgical outcome predictions, provided by the prediction module (116), include expected surgery duration, estimated blood loss during surgery along with contributing risk factors graph, and risk factors for post-operative ward placement.

8. A method for predicting surgical outcomes through an artificial intelligence (Al) based system (100), comprising: collecting, by a data collection module (122), input data of a patient from various sources such as electronic medical records (EMRs), Data repositories and Databases through a data collection module 122;standardizing and pre-processing, by a standardization and pre-processing module (124), the collected input data to ensure integrity and compatibility based on system's processing requirements and obtain pre-processed data; transmitting, by a API module (126), the pre-processed data from the standardization and pre-processing module (124) to a server (104) via a network (102), subsequently transmitting the pre-processed data to an input module (112) through the server (104); enabling, by the input module (112), the Al-based system (100) initiates authentication and validation of the pre-processed data related to one or more personal details, vital signs data, current medical history, anesthesia assessment data, laboratory parameters, and surgical procedure details, thereby obtaining validated data of the patient; standardizing and modifying the validated data using a processing module (114) for subsequent analysis; utilizing machine learning models within a prediction module (116) to generate key surgical outcomes based on the standardized data; providing a recommended protocol of subsequent actions for patients, based on the predicted surgical outcomes, using a clinical pathway module (118), wherein the recommended protocol of subsequent actions includes specifying requirements for blood and blood-related products; and presenting predicted surgery duration, blood loss estimates, contributing risk factors, and postoperative patient placement using a risk prediction response module (120), and generating a PDF report containing these predictions.6, DATE AND SIGNATURE:Dated this 28thday of August, 2024PATENT AGENT NAME: VARALAKSHMI VANAMINPA - 3287

Images