Systems and methods for optimizing medical interventions using predictive models

Abstract

A computer-implemented method for prescribing optimized medical interventions includes searching an updated electronic medical record (EMR) of a patient and mapping the patient's diagnosis to a medical treatment database to select a plurality of possible medical intervention options based on a score exceeding a predefined threshold score. The method includes determining a rank order of the selected plurality of medical intervention options by comparing a simulation result for each option performed by a medical prediction algorithm against a respective choice among each of the selected possible medical interventions. The method also includes the patient's physician or the patient's electronic medical record database receiving the rank order of the recommended medical intervention options including the possible options and associated metrics based on an accepted level of the simulated results.

Description

[Technical field]

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 197,807, filed June 7, 2021, which is incorporated by reference in its entirety.

[0002] The present disclosure relates generally to systems and methods for optimizing medical interventions using predictive models, and more particularly, to systems and methods for optimizing medical therapy decisions based on predictive model-based guidelines. [Background technology]

[0003] Medical therapy involves decisions made by clinicians, such as physicians, when evaluating a patient. It is not surprising that an incorrect decision made by a physician can mean the life or death of the patient. An incorrect (or suboptimal) decision may include failure to diagnose or misdiagnosis, or selection of a treatment plan option from multiple options available that may lead the patient and physician down a path that results in a suboptimal outcome. The primary stakeholder is the patient, who wants the longest life expectancy and the best quality of life (at a specific cost or at the lowest cost or affordable cost). The secondary stakeholders are the treating clinicians and institutions / hospitals, who want to ethically maximize profits and reputation.

[0004] For many years, medical societies have provided, created, and updated treatment guidelines for almost every prevalent disease based on population-level experience with each prevalent disease or when there are promising and effective new treatments that become available, but it is well known that these guidelines may not necessarily provide optimal outcomes for every individual patient. Typically, the treatment options provided in these guidelines may involve shared decision-making, involving the patient in the decision-making process. As population-guided guidelines are routine, current medical practice cannot take into account all the details of a patient's specific genetic, biological, anatomical, and / or physiological characteristics according to the guidelines, which cannot guarantee the most optimal outcome for an individual patient. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] US Patent Application Publication No. 2019 / 0298450 Summary of the Invention [Means for solving the problem]

[0006] The computer-implemented method for prescribing an optimized medical intervention includes executing, by at least one processor in the server, a medical prediction algorithm, stored in a memory, for prescribing an optimized medical intervention according to a diagnosis.

[0007] The system for prescribing an optimized medical intervention includes at least one processor in a server that executes a medical prediction algorithm, stored in a memory, for prescribing an optimized medical intervention according to a diagnosis.

[0008] The non-transitory computer readable medium stores in a memory program code of a medical prediction algorithm, which when executed by at least one processor of the machine, causes the machine to perform steps for prescribing an optimized medical intervention.

[0009] Prescribing an optimized medical intervention includes retrieving the patient's updated electronic medical record (EMR) from a patient database, the electronic medical record including at least two or more of the patient's demographic data, medical symptoms, vital signs, medications, surgical history, family medical history, genetic data, laboratory test data, disease records, allergies, x-ray or computer generated tomography images, and medical insurance information.

[0010] Prescribing an optimized medical intervention includes mapping the patient's diagnosis to a medical treatment database to select a plurality of possible medical intervention options based on a score exceeding a predefined threshold score, the score being a linear or non-linear combination of individual scores assigned to each of a plurality of objective functions including at least one or more of treatment duration, total treatment cost, treatment risk factors, planned life expectancy of treatment, treatment success rate, rehabilitation duration, outpatient rehabilitation cost, quality of life index after treatment, implant device useful life, equipment rating, and reimbursement cost from insurance providers.

[0011] Prescribing the optimized medical intervention includes determining a rank order of the selected medical intervention options by comparing simulation results for each option performed by the medical prediction algorithm for a respective choice among each of the selected possible medical interventions. Prescribing the optimized medical intervention includes the patient's physician or the patient's electronic medical record database receiving the rank order of the recommended medical intervention options including possible options and associated metrics based on an accepted level of simulated outcomes. [Brief description of the drawings]

[0012] [Figure 1A] Figure 1 shows an example illustration of the use of Artificial Intelligence / Machine Learning (AI / ML) algorithms to predict healthcare implications for both patients and hospitals for each decision tree / pathway, including life expectancy, cost and quality of life indices for each decision path. [Figure 1B] Figure 14 shows an example illustration of the use of artificial intelligence / machine learning (AI / ML) algorithms to predict healthcare implications for both patients and hospitals for each decision tree / pathway, including life expectancy, cost and quality of life indices for each decision path. [Diagram 2]FIG. 1 illustrates an example method for using an artificial intelligence driven guidance system to develop an optimal treatment pathway for a patient requiring heart valve replacement. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] The present disclosure provides an additional step to personalize population guidance guidelines by considering patient-specific information, which may be a prediction of known risks to treatment options available to the patient. The advent of predictive models using some combination of artificial intelligence, machine learning, big-data, and computational simulations offers the possibility that, once these models begin to take into account the specific genetic, biological, anatomical, and / or physiological characteristics of the patient, they may be able to predict, with some certainty (or uncertainty), the likely course of events for the patient over time.

[0014] The present disclosure herein may be a system for optimizing decision making for structural cardiac therapies such as valve repair and replacement. Those skilled in the art will recognize that the system may be configured to work with other cardiac or non-cardiac surgical therapies requiring surgery or catheter-based intervention, including robotic surgery or intervention.

[0015] The system can also be generalized to the treatment of any disease for which multiple treatment options exist.

[0016] Transcatheter valve repair or replacement (TVR), for example for the aortic, mitral, pulmonary and tricuspid valves, can provide treatment to patients with significant valve disease who cannot undergo traditional open heart surgery, and to high-risk patients with a variety of comorbidities. Despite the benefits associated with TVR, complications can still occur, such as conduction abnormalities, significant residual regurgitation or leakage, tissue damage, valvular thrombosis, valvular embolism and flow disturbances, as well as other cerebrovascular events. Complication events can include coronary injury, perivalvular regurgitation and thrombosis. Other complications include complications arising from the delivery process, such as vascular injury, or the inability of the delivery catheter to navigate in narrowing vessels. Unlike more invasive procedures such as open heart surgery, the decision to subject a patient to TVR is complex and can depend on many factors. These factors include age, sex, medical conditions, frailty, the patient's Society of Thoracic Surgeons (STS) risk score, prior history related to structural heart procedures, the complexity of the catheter delivery route, and the patient's ability to recover from open heart surgery.

[0017] Current guidelines by the American Heart Association and the American College of Cardiology or other associations recommending treatment strategies for heart valve repair and replacement simply rely on population-based risk assessments that, in some cases, ignore relevant patient-specific factors such as genetic, biological, anatomical and / or physiological characteristics of the lesion or multiple mixed lesions afflicting various cardiac structures. Risk calculators such as the STS Predicted Risk of Mortality (PROM) score are often used, but these calculators do not take into account the genetic, biological, anatomical and / or physiological characteristics of the lesion or mixed lesions. Risk calculators typically stand on statistical models based on national registry data compiled across all sites performing a particular procedure. Prediction capacity is limited to the recorded parameters, which typically include only general parameters such as age, sex, and other medical factors such as comorbidities, basic life and metabolic parameters (e.g., kidney function, lung function, etc.). Detailed anatomical information is not usually included in these models.

[0018] In an example, computational models can be utilized to simulate a transcatheter valve replacement or repair procedure and potentially assess the risk of some adverse outcomes (e.g., valve thrombosis) that may occur during surgery or even months after surgery. Additionally, these models can be used to predict future risks of interventions when current interventional devices fail (e.g., future valves within valves, i.e., TAVR within TAVR procedures). Examples of these models can be found in U.S. Patent Application Publication No. 2019 / 0298450, entitled "SYSTEMS AND METHODS FOR PREDICTIVE HEART VALVE SIMULATION," which describes a method for predicting the gap size α in TAV stent development using a parametric analysis engine. 2D We model the distance, the gap size α between the tip of the coronary leaflet and the coronary ostium of the coronary artery. 2D The gap size α can be correlated with the coronary injury risk level.2D A summary of the clinical status of some patients who used the is compiled in Table 1.

[0019] [Table 1]

[0020] These simulations must begin, in some cases, with the generation of a 3D model from the patient's medical images, followed by subjecting the 3D model to a simulation of the TAV using a computational framework capable of predicting the structural response as well as the hemodynamic performance and other flow parameters and patterns. Thus, if one wishes to personalize the risk score, it is now possible to personalize the already developed risk score by adding individual level predictions to the calculation. By using a new risk calculator that combines population-based predictions with personalized simulations to predict specific adverse outcomes (of current and future interventions), a more accurate risk calculation is obtained at the individual level, thereby optimizing decision-making. The more personalized image, genetic data, or physiological data that is included in the personalized prediction of adverse outcomes, the more accurate the combined model will be in predicting the personalized risk for a given medical treatment or intervention.

[0021] While it can be useful to have computer simulations to sift out the individual adverse outcomes, optimal decisions for patients require a more holistic picture in which all options (invasive vs. transcatheter vs. medical management) are considered. For each option considered, it is important to identify the risk of the adverse outcome that may occur along with the potential implications of the various adverse outcomes for the patient as well as the treatment center (e.g., hospital). For the patient, the implications deal with life expectancy, quality of life, and costs. For the hospital, the implications result in the length of the patient's stay in hospital, and the additional costs that will be incurred by the hospital, as well as changes in reputation. In addition to considering these implications, another set of implications for the patient is that individual decisions may set the stage for other decisions to be made in the future. For example, if a decision to implant a prosthetic valve now is being considered, it should also be anticipated that the patient will need another valve within about 10 years. Furthermore, if a patient will require a coronary intervention in the future (based on prediction of coronary artery disease), current decisions based on the extent of progression of the patient's coronary artery disease as seen from computerized tomography (CT) scan images will necessarily also include coronary access as a determining factor.

[0022] This type of holistic view maps out early decisions from physicians or by cardiac team groups to outcomes for both patients and hospitals. As an illustration, unlike open heart surgery, a treatment decision for a younger, lower risk patient using TVR (based only on STS score) may mean certainty of future interventions such as additional transcatheter or open heart surgery in the future. It is similar to a game of chess where players can consider future moves and where opposing moves have a better probability of winning. Similarly, medical decision support systems that lay out in a quantitative way all the implications from individual decision points to select the best decision path for the best outcome are also important. Such optimal decisions cannot be made without system level planning and intelligence driven guidance systems. As explained above, such systems collect data in near real time from all centers, which includes genetic, anatomical and other physiological parameters in the data collection, updating population-based predictive models with increasing accuracy. The system also has the same simulation capability to simulate the biomechanical interaction of the device in the patient while considering all possible treatment options (current and future predicted interventions) and then calculate an accurate risk score for each treatment pathway. The system can be configured to visually interactively display the decision tree or map on a display (e.g., a touch screen device or a virtual reality display device or an augmented reality device) and highlight the most optimal decision pathway or paths for each of the optimization variables (e.g., cost, patient lifespan, etc.). Such a system will act as an intelligence and will provide justification for the best option for the clinician to consult with his team and the patient and family for the implementation of the overall selected treatment plan. Such a system, which can be run centrally using a service or a cloud computing platform, is the subject of this disclosure. Furthermore, every patient can install an application on his personal device connected to the system and the system will display the patient's life expectancy in real time, which can be updated every time the patient's electronic medical record changes.Additionally, the application can also be connected to wearable sensors that can also provide updated biological, physiological, cognitive, mental and other biomarker data that will also be used in the system's predictive algorithms. Such applications would be useful for patients to adjust their lifestyle to maximize their life expectancy and quality of life.

[0023] In the present disclosure, machine learning and / or artificial neural network or deep learning algorithms are trained and updated at regular time intervals to track patient health care parameters and outcomes such as length of stay, cost of care, itemized list of supplies used during care (canes, saline bags, accessories, etc.), valve durability, life expectancy, and probability of complications from national, regional to center / hospital specific levels. For example, machine learning algorithms could predict length of stay, cost, and projected life expectancy for patients who have had TAVR performed and ended up having permanent pacemakers installed. These algorithms could be trained on retrospective data from all available medical data for patients who have had structural heart procedures such as open heart surgery or transcatheter procedures.

[0024] These AI / ML algorithms are then combined with computational predictive models to form a system optimization methodology or software to model all possible options, current and future, across all possible time courses for a patient. For example, a patient diagnosed with aortic stenosis will be offered all options, such as open heart surgery or transcatheter valve replacement. For each option, the computational simulation will predict the risk of adverse outcomes with the individual devices available for the option. The computational simulation can include Monte-Carlo simulation to incorporate uncertainty while deploying the device to predict the probability of adverse outcomes. For each device selected, future valves in valve device simulation will also be performed in a similar manner. The simulation can also include predictive remodeling of the heart structure (learned from AI / ML training from data collection), as well as predictive increases in other structural heart diseases, such as coronary artery disease. For each decision tree / pathway, the AI / ML algorithm will predict the healthcare implications for both the patient and the hospital, including life expectancy, cost, and quality of life index for each decision path. This concept can be illustrated in Figures 1A and 1B.

[0025] The system can display decision pathways as a visual map highlighting best and worst pathways, including intermediate pathways. Clicking on a pathway will display parameters important to the hospital, such as life expectancy, cost, quality of life, and potential cost savings and profitability.

[0026] Some example steps performed by an artificial intelligence driven guidance system are shown in Figure 2. In particular, Figure 2 illustrates the construction of an optimization system using an example to develop an optimal treatment pathway for a patient in need of a heart valve replacement.

[0027] In the illustrative example, a female patient in her 50s suffers from aortic stenosis with a calcified existing bioprosthetic valve. An (Artificial Neural Network) ANN model having at least one layer and at least one neuron is trained to output the life expectancy of an individual based on their current heart condition. The training dataset may include a database of the patient's electronic medical history including various conditions and medical health parameters such as the severity of each condition. The database may span the maximum possible range, preferably 10 years or more. The model may be trained weekly upon syncing with the medical institution where new data is added. The model may not require any personal identifiers. However, the model may include relevant information including genetics, race, sex, geographic location, and relevant functional status such as kidney, lung, liver, blood cholesterol, etc. Family history and all past diagnoses and their treatment status. The ANN model may output the life expectancy in the form of predicted age at death and years of life remaining. The model may also output the life expectancy when the disease status is hidden from the model. This may mean that the model may assume that there is no current disease and that the output of the model may be the life expectancy in the absence of new diagnoses.

[0028] This trained predictive life expectancy model may predict only 1 year if untreated. The algorithm may also predict the same patient without aortic stenosis to have a life expectancy of 85 years. The system may recognize that a possible best case scenario for developing a treatment pathway is to attempt to add 34 years to the expected life expectancy. The system may set a goal to select a treatment pathway that will result in a life expectancy between 51 and 85 years (i.e., adding 34 additional expected years).

[0029] The system has another ANN model trained to output the number of years a valve will last before failure. This model is similar to the ANN model already described and is trained on the medical records of all patients who have received replacement valves. The model can be trained to output the predicted useful life of all valves on the market based on the patient's age and any vital parameters including but not limited to renal function.

[0030] For patient examples, the model can project the new Sapien valve available from Edwards Lifesciences (hereinafter "Sapien") to last 10 years, while the new Evolut valve (e.g., Medtronic Evolut™ TAVR, hereinafter "Evolut") can last 6 years. The model can also project the new Carpentier Edwards Perimount valve (hereinafter "Perimount") to last 16 years, the Magna valve to last 14 years, and the Mitroflow (e.g., Mitroflow aortic valve) to last 12 years.

[0031] The next step in the system may be a treatment pathway generator. This may be an algorithm that considers the available options and lays out all possible treatment pathways, including medical management only pathways. These treatment pathways may also be called interventions. The algorithm ensures that the treatment pathways, or intervention choices, are laid out such that the combined useful life of all of the devices in the pathway may be longer than the longest possible life expectancy of the patient. Interventions may also include variations such as different depths of implantation of the valve, different sizing (oversizing or undersizing by a set amount), different angles of deployment due to changes in how the guidewire can place the implant in the aortic root, and so on.

[0032] In the illustrative example, the following intervention selections can be generated: Choice No. 1: Medical Management. Option No. 2: Direct heart valve replacement with Perimount, followed by direct heart valve replacement with Perimount, followed by direct valve replacement with Perimount = total useful life 16+16+16=48 years. Option No. 3: Open heart valve replacement with Perimount, followed by TAVR valve-in-valve (ViV) replacement with Evolut, followed by valve-in-valve-in-valve (ViViV) replacement with Sapien = total useful life 16+6+10=32 years.

[0033] All medical interventions can then be passed through the following algorithm, which evaluates the overall probability of a life expectancy of 34 years or more for each medical intervention, and can sort the choices from most likely to least likely.

[0034] The algorithm can include information from training based on data from participating institutions / hospitals, etc., and this information can be updated as new data becomes available. The algorithm, which has a pre-calculated probability distribution function of life expectancy after open heart surgery, can be conditioned on the patient's selected valve, geographic location, patient's age at time of surgery, comorbidities, and other medical information (current and planned for the future). Another condition is the absence of adverse outcomes.

[0035] The algorithm, with its pre-calculated probability distribution function of life expectancy after transcatheter valve replacement, can be conditioned on the patient, the selected valve, geographic location, age at time of surgery, comorbidities, and other medical information (present and projected for the future). Another condition is the absence of adverse outcomes.

[0036] An algorithm with pre-computed probability distribution functions of life expectancy conditioned on each adverse outcome (and any combination) may result from at least root fracture, perivalvular regurgitation level, patient-prosthesis mismatch, valve thrombosis, coronary lesions, valve embolism, permanent pacemaker placement, pulsation, combinations (e.g., patient-prosthesis mismatch + valve thrombosis), etc.

[0037] The algorithm can calculate the probability of each adverse outcome occurring at each step in the treatment pathway by running a biomechanical prediction simulation based on the patient's medical images. The individual treatment pathway can be simulated. The biomechanical prediction model can include algorithms that can be used to (1) predict the likelihood of thrombosis after transcatheter aortic valve (TAV) replacement (TAVR) based on valve and patient-specific anatomical and hemodynamic parameters, and (2) select the optimal valve and placement for an individual patient to minimize the likelihood of thrombosis. The algorithm can be designed to incorporate valve and patient-specific parameters with a degree of empirical and quasi-empirical modeling to rapidly predict areas of blood flow stagnation or the degree of blood flow stagnation that may lead to blood clot formation near the prosthetic valve, and then output the probability of each adverse outcome occurring at each step in the treatment pathway.

[0038] The algorithm can run Monte Carlo simulations for each medical intervention given the individual probabilities of each adverse outcome occurring for each treatment pathway using probability density functions and conditional probability density functions to calculate the total probability function of life expectancy for each treatment pathway. These Monte Carlo type simulations can be driven by a random number generator following the given probability density functions and probabilities.

[0039] The algorithm can use the final probability density function to predict the probability of a life expectancy of more than 34 years. For example, the model can output medical interventions in order from most likely to least likely.

[0040] Another algorithm can display length of stay, projected hospital costs, reimbursable costs and non-reimbursable costs per care pathway. This algorithm can be an ANN model trained on data that can include length of stay and cost information from participating hospitals, etc.

[0041] The above description is illustrative. Of course, it is not possible to describe every conceivable combination of components or methods, but one of ordinary skill in the art will recognize that many further combinations and permutations are possible. Accordingly, the present disclosure is intended to encompass all such alternatives, modifications, and variations that are within the scope of the present application, including the appended claims. Furthermore, when the present disclosure or claims recite an element in the "singular," a "first" element, or an "other" element, or equivalents thereof, it should be construed as including one or more elements other than one such element, even if it does not require or exclude two or more such elements. As used herein, the term "comprises" means including but not limited to, and the term "comprising" means including but not limited to. The term "based on" means based at least in part on.

Claims

1. A computer-implemented method for prescribing an optimal treatment pathway, comprising: retrieving data relevant to the diagnosis of the disease, the data including medical images and the patient's updated electronic medical record (EMR) including at least one of the patient's demographic data, pathological symptoms, vital signs, medications, surgical history, family medical history, genetic data, laboratory test data, disease records, allergies, and medical insurance information; generating available treatment pathway options including at least one surgical procedure or intervention; using at least one model to predict at least one implication for each of the available treatment pathway options based on the data and a simulated adverse outcome of the at least one surgical treatment or intervention, the adverse outcome being simulated by a predictive model including a biomechanical model based on the treatment images; interactively updating and displaying a decision tree including the available treatment pathway options, each having at least one corresponding objective function, wherein the at least one model includes a predictive life expectancy model trained to predict the at least one objective function including a predicted life expectancy; and 4. A computer-implemented method comprising:

2. The computer-implemented method of claim 1, wherein the at least one model further includes an artificial intelligence / machine learning algorithm that predicts the at least one implication.

3. The computer-implemented method of claim 1, wherein the at least one model further comprises a statistical model that predicts the at least one implication.

4. The computer-implemented method of claim 1, wherein the at least one model further comprises a quasi-experimental or experimental model that predicts the at least one implication.

5. The computer-implemented method of claim 1, wherein the at least one model further comprises a reduced order model that predicts the at least one implication.

6. The computer-implemented method of claim 1, wherein the at least one objective function further includes at least one of the following: duration of treatment, total cost of treatment, risk factors of treatment, planned life expectancy of treatment, success rate of treatment, duration of rehabilitation, outpatient rehabilitation costs, quality of life index after treatment, implant device useful life, equipment rating, and reimbursable costs from an insurer.

7. The computer-implemented method of claim 1, further comprising interactively updating and displaying the decision tree as new data is added to the patient's EMR.

8. The computer-implemented method of claim 1, further comprising the step of interactively updating and displaying the decision tree having the available treatment pathway options in ranked order based on a selected objective function.

9. The computer-implemented method of claim 1, further comprising the step of performing a Monte-Carlo simulation using a probability density function of the adverse outcomes to calculate an aggregate probability function of life expectancy for each of the available treatment pathway options, the probability density function including pre-computed probability density functions of individual probabilities of each of the adverse outcomes and life expectancy after each of the at least one surgical treatment or intervention.

10. The computer-implemented method of claim 1, wherein the medical images include X-ray images and / or computer-generated tomography images.

11. A system for prescribing an optimal treatment pathway, comprising: at least one memory containing instructions; at least one processor configured to execute the instructions; which, when the instructions are executed, retrieving data relevant to the diagnosis of the disease, the data including medical images and the patient's updated electronic medical record (EMR) including at least one of the patient's demographic data, pathological symptoms, vital signs, medications, surgical history, family medical history, genetic data, laboratory test data, disease records, allergies, and medical insurance information; generating available treatment pathway options including at least one surgical procedure or intervention; using at least one model to predict at least one implication for each of the available treatment pathway options based on the data and a simulated adverse outcome of the at least one surgical treatment or intervention, the adverse outcome being simulated by a predictive model including a biomechanical model based on the treatment images; interactively updating and displaying a decision tree including the available treatment pathway options, each having at least one corresponding objective function, wherein the at least one model includes a predictive life expectancy model trained to predict the at least one objective function including a predicted life expectancy; and on the at least one processor.

12. The system described in claim 11, wherein the at least one model further includes an artificial intelligence / machine learning algorithm that predicts the at least one implication.

13. The system of claim 11, wherein the at least one model further comprises a statistical model that predicts the at least one implication.

14. The system of claim 11, wherein the at least one model further comprises a quasi-experimental or experimental model that predicts the at least one implication.

15. The system of claim 11, wherein the at least one model further includes a reduced order model that predicts the at least one implication.

16. The system described in claim 11, wherein the at least one objective function further includes at least one of the following: duration of treatment, total cost of treatment, risk factors of treatment, planned life expectancy of treatment, success rate of treatment, duration of rehabilitation, outpatient rehabilitation costs, quality of life index after treatment, implant device useful life, equipment rating, and reimbursable costs from insurers.

17. The system of claim 11, wherein when the instructions are executed, the at least one processor is caused to perform the step of interactively updating and displaying the decision tree as new data is added to the patient's EMR.

18. The system described in claim 11, wherein when the instructions are executed, the at least one processor is caused to perform the step of interactively updating and displaying the decision tree having the available treatment pathway options in a ranked order based on a selected objective function.

19. The system of claim 11, wherein when the instructions are executed, the system causes the at least one processor to perform the following steps: performing a Monte-Carlo simulation using a probability density function of the adverse outcomes to calculate an aggregate probability function of life expectancy for each of the available treatment pathway options, the probability density function including pre-calculated probability density functions of individual probabilities of each of the adverse outcomes and life expectancy after each of the at least one surgical treatment or intervention.

20. The system of claim 11 , wherein the medical images include X-ray images and / or computed tomography images.

Images