Model-based supervised machine learning system for cancer detection from clinical datasets

Abstract

A supervised, machine learning-based cancer detection system (100) for analyzing clinical data sets and predicting cancer in a patient, the system comprising: a data collection module (1) configured to collect anonymized patient data from one or more clinical data sources including structured demographic data, diagnostic reports, pathological results and genetic markers; a data preprocessing module (2) configured to clean the acquired data by performing imputation of missing values, detection and treatment of outliers, normalization and coding of categorical variables; a feature selection and engineering module (3) configured to apply domain-specific feature selection techniques, including mutual information analysis, correlation analysis and recursive feature elimination, and construct derived composite features, including tumor progression rate or diagnostic scores; a machine learning model training module (4) configured to train one or more supervised learning algorithms, including Random Forest, Support Vector Machine (SVM), Gradient Boosting Machine (XGBoost) or Deep Neural Networks, using the processed and featured clinical data; a model evaluation module (5) configured to evaluate the trained model using metrics such as accuracy, precision, recall, F1 score and AUC-ROC with cross-validation techniques for performance validation; an interpretability module (6) configured to generate feature meanings and model explanations using interpretability tools such as SHAP (SHapley Additive Explanations) or LIME (Local Interpretable Model-Agnostic Explanations); a user interface module (7) comprising a dashboard or web-based interface configured to accept clinical inputs in real time and display the cancer prediction results along with interpretable rationales; a compliance and security module (8) configured to implement role-based access control, data encryption, and compliance with HIPAA and GDPR standards for data protection and patient confidentiality; a clinical feedback and integration module (9) configured to integrate the system with hospital information systems (HIS) and receive feedback from clinicians for continuous model improvement and retraining; and a deployment module (10) configured to deploy the trained and validated machine learning model on a secure cloud server or edge computing platform based on the availability of the healthcare infrastructure.

Description

FIELD OF THE INVENTION

[0001] The present invention relates to the field of medical diagnostics and artificial intelligence. More specifically, it concerns a supervised, machine learning-based system for cancer detection using structured clinical data sets. BACKGROUND OF THE INVENTION

[0002] The subject matter discussed in the "Background" section should not be assumed to be prior art merely by virtue of its mention in the "Background" section. Likewise, it should not be assumed that a problem mentioned in the Background section or related to the subject matter of the Background section has already been recognized in the prior art. The subject matter of the Background section merely presents various approaches, which may also be inventions in their own right.

[0003] Cancer remains a leading cause of morbidity and mortality worldwide, and early detection remains the most important factor in improving survival rates and treatment outcomes. Traditionally, the diagnostic workflow for cancer involves a combination of clinical assessments, radiological imaging, pathological evaluations, and laboratory tests. These diagnostic procedures are typically performed manually by healthcare professionals, who rely heavily on their expertise and experience to interpret complex and diverse patient data.

[0004] However, manual interpretation presents several challenges, including subjectivity, interobserver variability, delayed diagnoses, and limited scalability. Furthermore, with the increasing amount of available clinical data, particularly with the advent of electronic health records (EHRs), genomic sequencing, and digital pathology, it is becoming increasingly difficult for clinicians to process and analyze this data efficiently and consistently.

[0005] In recent years, machine learning (ML) has proven to be a powerful tool for detecting patterns in large data sets and making predictions. Supervised learning algorithms, in particular, have shown great promise in medical diagnostics, as they learn from labeled data to predict specific outcomes such as the presence, severity, or progression of a disease. These models have demonstrated their capabilities in classification and regression tasks with high-dimensional data.

[0006] Despite these advances, several challenges remain that prevent the widespread application of machine learning in clinical settings. First, raw clinical data is often noisy, incomplete, and heterogeneous, requiring careful preprocessing and normalization. Second, the process of selecting relevant features from the data is both critical and complex, as it directly impacts model performance and generalizability. Third, many machine learning models operate as "black boxes," making it difficult for healthcare providers to understand the reasons behind the predictions, thus limiting confidence and clinical usability. Finally, strict regulatory requirements and the need for ethical, secure, and compliant handling of patient data pose additional hurdles to practical use.

[0007] The present invention aims to overcome these limitations by providing a supervised, machine learning-based modeling system specifically tailored for cancer detection using structured clinical datasets. The system is designed to systematically process patient data—including demographic data, symptoms, diagnostic reports, laboratory values, pathology results, and genetic markers—through a pipeline that includes data cleaning, feature engineering, model training, scoring, and interpretability. The invention not only delivers high diagnostic accuracy but also emphasizes transparency, interpretability, and clinical integration.

[0008] By leveraging interpretation tools such as SHAP and LIME, the system provides explanations for its predictions, supporting evidence-based decision-making. Furthermore, it complies with healthcare regulations such as HIPAA and GDPR to ensure secure and lawful data handling. In this way, the invention bridges the gap between advanced AI technology and real-world clinical application, providing a robust, scalable, and ethical solution to support early and accurate cancer diagnosis.

[0009] The information disclosed in this Background section is provided solely to facilitate an understanding of the background of the invention and may therefore contain information that is not prior art and that is already known to a person of ordinary skill in the art in this country. SUMMARY

[0010] Before describing the present systems and methods, it should be noted that this application is not limited to the systems and methods described, as there may be several possible embodiments not expressly shown in this disclosure. It should also be understood that the terminology used in the description is for the purpose of describing the particular versions or embodiments only and is not intended to limit the scope of the present application.

[0011] The present invention provides a supervised machine learning-based system for cancer detection using clinical data sets.

[0012] The present invention discloses a supervised machine learning-based cancer detection system specifically designed for analyzing structured clinical datasets, supporting healthcare professionals in the early and accurate diagnosis of cancer. The system utilizes a modular architecture consisting of interconnected functional units that cover the entire lifecycle of data-driven cancer prediction—from data acquisition to interpretability, clinical feedback, and safe deployment.

[0013] The system includes a data acquisition module that enables the capture of anonymized clinical data from electronic health records (EHRs), hospital databases, or open biomedical repositories. These datasets include various patient characteristics such as age, gender, symptoms, tumor size, biopsy results, blood test results, and genetic markers.

[0014] After data input, a data preprocessing module performs essential operations, including imputing missing values, normalizing numeric variables, and coding categorical features, to ensure the data is clean, consistent, and model-ready. The processed data is then passed to a feature selection and development module, which applies statistical methods and machine learning-based techniques—such as correlation analysis, mutual information, and recursive feature elimination—to identify and retain the most clinically relevant variables. Furthermore, this module is capable of creating composite diagnostic features, such as tumor progression scores or risk indices, derived from multiple raw data sets.

[0015] The model training module uses one or more supervised machine learning algorithms—such as Random Forest, Support Vector Machine (SVM), XGBoost, or Deep Neural Networks—to train classifiers that can predict cancer status (e.g., benign vs. malignant) or cancer type (e.g., lung, breast, prostate). The models are validated with robust evaluation metrics such as accuracy, precision, recall, F1 score, and AUC-ROC, and refined with cross-validation techniques to ensure generalizability across populations.

[0016] To address the critical need for explainability in medical decision-making, the system integrates an interpretability layer with SHAP (SHapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations), which provide case-specific justifications for each prediction. These explanations shed light on the most influential clinical parameters underlying the model's decision and allow clinicians to trust, validate, and investigate the predictions.

[0017] A user interface module in the form of a web-based or desktop application allows healthcare providers to enter new patient data in real time, obtain predictions, and visualize the explanation results. The interface enables interactive exploration of prediction sensitivity by adjusting input variables and verifying outcome shifts, thus supporting decision-making and patient counseling.

[0018] The invention also addresses the legal and ethical aspects of using artificial intelligence in healthcare. The security and compliance module ensures full compliance with data protection laws such as the Health Insurance Portability and Accountability Act (HIPAA) and the General Data Protection Regulation (GDPR). This includes secure encryption, access control, audit trails, and anonymization of sensitive patient data.

[0019] To support real-world deployment and continuous improvement, the invention includes a clinical integration and feedback module that enables seamless integration with hospital information systems (HIS) via standardized APIs. Furthermore, it supports the collection of feedback from clinicians for continuous model improvement and bias reduction. A deployment module enables flexible hosting of the trained model either on a secure cloud infrastructure or on local devices, depending on the healthcare facility's resources. BRIEF DESCRIPTION OF THE DRAWING

[0020] To clarify various aspects of some embodiments of the present invention, a more detailed description of the invention will be given by reference to specific embodiments illustrated in the accompanying drawings. It is understood that these drawings show only illustrative embodiments of the invention and are therefore not to be considered limiting its scope. The invention will be described and explained with additional specificity and detail through the use of the accompanying drawings.

[0021] In order that the advantages of the present invention may be readily understood, a detailed description of the invention is discussed below in conjunction with the accompanying drawings, which, however, should not be construed as limiting the scope of the invention to the accompanying drawings, in which: Fig. shows the block diagram of a supervised machine learning-based system (100) for cancer detection from clinical data sets. DETAILED DESCRIPTION

[0022] The present invention relates to a supervised, machine learning-based model system (100) for cancer detection using clinical data sets.

[0023] Fig. shows the block diagram of a supervised machine learning-based system (100) for cancer detection from clinical data sets.

[0024] The present invention discloses a supervised, machine learning-based model system

[100] for cancer detection using clinical datasets. The system is designed to assist healthcare professionals in diagnostic decision-making by analyzing structured patient data and predicting the presence or type of cancer with interpretability and high accuracy. The system consists of several interdependent modules, each performing a specific function in the overall prediction workflow.

[0025] The cancer detection system (100) includes the following components: [1] Data acquisition module: Collects structured and anonymized clinical data from sources such as hospital databases or public repositories such as The Cancer Genome Atlas (TCGA) or SEER. Data may include age, gender, symptoms, biopsy results, laboratory results, tumor size, and genetic markers. [2] Data preprocessing module: Performs imputation of missing values, outlier detection and treatment, normalization, and coding of categorical variables (e.g., one-hot or label coding) to ensure that the data is suitable for machine learning. [3] Feature selection and development module: Apply statistical and machine learning techniques such as correlation analysis, mutual information, and recursive feature elimination to obtain relevant attributes. Composite features such as tumor progression rate or diagnostic risk scores can be created using clinical knowledge. [4] Training module for machine learning models: Trains one or more supervised learning algorithms such as Random Forest, Support Vector Machine (SVM), XGBoost, or Deep Neural Networks. The trained model classifies the clinical data into binary or multi-class labels such as benign / malignant or cancer types. [5] Model evaluation module: Evaluates the trained model using cross-validation and performance metrics such as accuracy, precision, recall, F1 score, and AUC-ROC to ensure reliability and generalization. [6] Interpretability module: Generates feature-level explanations using SHAP or LIME to reveal the most influential attributes for each prediction. This allows clinicians to understand and trust the model's decisions. [7] User interface module: Provides a dashboard or web-based interface for clinicians to enter patient data, obtain predictions, and visualize the model's interpretation. Users can simulate different inputs to observe variations in results. [8] Conformity and Safety Module: Implements data encryption, role-based access control, and compliance with standards such as HIPAA and GDPR for secure and compliant handling of clinical data. [9] Clinical Feedback and Integration Module: Supports integration with hospital information systems (HIS) and incorporates clinician feedback for regular model training and performance improvements.

[10] Deployment module: The fully trained model is deployed on secure cloud platforms or edge devices, depending on infrastructure availability. Remote update mechanisms enable regular improvements and patches.

[0026] The data collection module [1] is configured to collect anonymized, structured clinical data from various sources, including electronic health records (EHRs) and publicly accessible repositories such as The Cancer Genome Atlas (TCGA), Surveillance, Epidemiology, and End Results Program (SEER), and other institutional biobanks. Clinical data includes demographics (e.g., age and gender), clinical symptoms, pathology reports (e.g., tumor size and biopsy results), laboratory test values (including blood counts and tumor markers), comorbidities, and genomic or molecular profile data, where available. The data is stored in structured formats to facilitate downstream machine learning processing.

[0027] The data preprocessing module [2] receives the raw clinical data and performs rigorous cleaning operations. Missing values are corrected using statistically appropriate methods such as mean, median, or model-based imputation. Outliers are detected using statistical thresholds or clustering techniques and treated with truncation or smoothing. Categorical variables (e.g., cancer stage or patient sex) are encoded using label coding or one-hot coding. Continuous features are normalized or standardized to ensure a consistent scale and model compatibility. This module ensures machine-readable data and reduces the risk of bias or errors during model training.

[0028] The feature selection and development module [3] is responsible for extracting the most informative and clinically relevant features from the preprocessed data. Statistical techniques such as Pearson / Spearman correlation coefficients, mutual information values, and recursive feature elimination (RFE) are used to select the features with the highest predictive power. In addition to the original clinical parameters, new composite features are developed. For example, the tumor progression rate can be derived from changes in tumor size over time, and diagnostic risk scores can be formulated by integrating multiple biomarkers into a single indicator. These developed features improve the model's ability to generalize and capture domain-specific insights.

[0029] The module for training machine learning models [4] includes a library of supervised learning algorithms that are trained using the selected and developed features. Algorithms such as Random Forest, Support Vector Machines (SVM), Extreme Gradient Boosting (XGBoost), and Deep Neural Networks are implemented depending on the complexity and nature of the clinical dataset. The system supports both binary classification (e.g., benign vs. malignant) and multiclass classification (e.g., breast, colon, lung, prostate cancer types). The model hyperparameters are optimized using grid search, random search, or Bayesian optimization techniques.

[0030] The model evaluation module [5] quantitatively evaluates the performance of the trained models using a combination of evaluation metrics, including accuracy, precision, recall, F1 score, and the area under the receiver operating characteristic curve (AUC-ROC). The models are validated using k-fold cross-validation and bootstrapping to ensure their robustness and reduce the risk of overfitting. The performance metrics are recorded and analyzed to select the best-performing model for deployment.

[0031] To increase transparency and promote clinician confidence, the system includes an interpretation module [6]. This module leverages explainable AI techniques such as SHAP (SHapley Additive Explanations) and LIME (Local Interpretable Model-Agnostic Explanations) to generate visual and textual explanations for each prediction. These explanations highlight the most influential features that contributed to the predicted outcome, allowing clinicians to understand the model's reasoning and compare it with their own knowledge.

[0032] The user interface module [7] is a front-end interface designed as a web-based dashboard or desktop application. It allows authorized users to enter new patient data in real time, obtain prediction results, and view feature importance charts and explanations. The interface supports parameter sensitivity analyses, allowing users to simulate changes in input values to observe the impact on the predicted cancer label. The dashboard is designed for user-friendliness and complies with medical software usability standards.

[0033] The Compliance and Security module [8] ensures that all patient data is handled in compliance with healthcare privacy regulations such as the Health Insurance Portability and Accountability Act (HIPAA) and the General Data Protection Regulation (GDPR). This module implements role-based access control, data anonymization, and end-to-end encryption for all data transfer and storage activities. Secure audit trails are maintained to track all system access and usage events.

[0034] The Clinical Feedback and Integration module [9] was developed to facilitate the system's integration with existing hospital information systems (HIS) and electronic medical records (EMR) via standardized APIs. Furthermore, the module provides a feedback mechanism that allows clinicians to comment on, validate, or challenge the model predictions. This feedback is logged and used during retraining cycles to improve model performance and reduce systematic bias.

[0035] Finally, the deployment module

[10] provides options for deploying the trained and validated model on various computing platforms. These include secure cloud servers for centralized hospital systems and edge devices for deployment in resource-constrained environments such as rural clinics. The module supports continuous learning and remote updates through version control and scheduled retraining cycles based on updated clinical data sets.

[0036] These modules work synergistically to deliver a robust, interpretable, secure, and clinically usable cancer detection system that supports healthcare providers in real-time diagnostic workflows.

Claims

[1] A supervised, machine learning-based cancer detection system (100) for analyzing clinical data sets and predicting cancer in a patient, the system comprising: a data collection module (1) configured to collect anonymized patient data from one or more clinical data sources including structured demographic data, diagnostic reports, pathological results and genetic markers; a data preprocessing module (2) configured to clean the acquired data by performing imputation of missing values, detection and treatment of outliers, normalization and coding of categorical variables; a feature selection and engineering module (3) configured to apply domain-specific feature selection techniques, including mutual information analysis, correlation analysis and recursive feature elimination, and construct derived composite features, including tumor progression rate or diagnostic scores; a machine learning model training module (4) configured to train one or more supervised learning algorithms, including Random Forest, Support Vector Machine (SVM), Gradient Boosting Machine (XGBoost) or Deep Neural Networks, using the processed and featured clinical data; a model evaluation module (5) configured to evaluate the trained model using metrics such as accuracy, precision, recall, F1 score and AUC-ROC with cross-validation techniques for performance validation; an interpretability module (6) configured to generate feature meanings and model explanations using interpretability tools such as SHAP (SHapley Additive Explanations) or LIME (Local Interpretable Model-Agnostic Explanations); a user interface module (7) comprising a dashboard or web-based interface configured to accept clinical inputs in real time and display the cancer prediction results along with interpretable rationales; a compliance and security module (8) configured to implement role-based access control, data encryption, and compliance with HIPAA and GDPR standards for data protection and patient confidentiality; a clinical feedback and integration module (9) configured to integrate the system with hospital information systems (HIS) and receive feedback from clinicians for continuous model improvement and retraining; and a deployment module (10) configured to deploy the trained and validated machine learning model on a secure cloud server or edge computing platform based on the availability of the healthcare infrastructure. [2] The system of claim 1, wherein the clinical data set includes information selected from the group consisting of: age, gender, symptoms, tumor size, biopsy result, blood test report, family history, and genomic mutation status. [3] The system of claim 1, wherein the feature selection and development module (3) is further configured to construct time-based history features and diagnostic risk scores using domain heuristics. [4] The system of claim 1, wherein the training module (4) of the machine learning model is configured to classify input data into binary (benign or malignant) or multi-class cancer labels selected from cancer types including, but not limited to, lung, breast, colon, liver, and prostate cancer. [5] The system of claim 1, wherein the interpretability module (6) generates visual representations of the feature contributions per prediction to improve clinical decision making and model confidence. [6] The system of claim 1, wherein the user interface module (7) enables clinicians to modify input parameters and observe variations in the prediction results for clinical experiments. [7] The system of claim 1, wherein the deployment module (10) includes mechanisms for periodic remote updating and retraining using updated clinical records and feedback data. [8] The system of claim 1, wherein the clinical feedback and integration module (9) comprises a feedback loop that allows clinicians to evaluate predictions and note erroneous results for retraining purposes. [9] The system of claim 1, wherein the compliance and security module (8) includes audit trails for monitoring access and usage history of patient data and prediction results. [10] The system of claim 1, wherein the trained model is validated on retrospective patient data and subjected to prospective pilot testing in clinical settings prior to full deployment.

Images