Ai-based smart health screening system for identifying individuals at risk of developing cancers, cardiovascular disease and type 2 diabetes

Abstract

The invention is an AI-driven health screening system for early detection of cancers, cardiovascular diseases, and type 2 diabetes. It integrates multiple screening methods, using machine learning, deep learning, and natural language processing to analyze genetic and health data. A federated learning framework ensures high prediction accuracy while maintaining data privacy. The system also employs a blockchain-based electronic health record (EHR) for secure data management. AI models, including neural networks and support vector machines, assess risk factors and provide personalized healthcare recommendations. Designed with a three-tier architecture, it supports deployment as a web service, software, or integration into existing programs. The system enhances early disease detection, optimizes healthcare resources, and improves patient outcomes by offering a scalable, efficient, and non-invasive screening solution.

Description

[0001] AI-BASED SMART HEALTH SCREENING SYSTEM FOR IDENTIFYING INDIVIDUALS AT RISK OF DEVELOPING CANCERS, CARDIOVASCULAR DISEASE AND TYPE 2 DIABETES

[0002] TECHNICAL FIELD

[0003] The present disclosure generally relates to the field of digital health, artificial intelligence, and predictive analytics in medical screening. More specifically, it pertains to an Al-driven platform designed for the early identification of individuals at risk of developing cancers — such as breast, prostate, colorectal, and cervical cancer, cardiovascular disease — as well as type 2 diabetes. The system integrates multiple screening modalities using advanced machine learning algorithms, deep learning models, and natural language processing techniques to assess genetic information and traditional risk factors. By leveraging Al-powered digital analysis, the disclosed invention enhances early disease detection, improves risk assessment accuracy, and optimizes healthcare resources. The innovative software framework employs a three-tier architecture, comprising a presentation layer for user interaction, a business logic layer for data processing, and a data access layer for managing system data. This holistic approach addresses current limitations in screening programs, improves accessibility to predictive diagnostics, and facilitates personalized recommendations for preventive healthcare strategies.

[0004] BACKGROUND

[0005] The identification of individuals at risk for developing cancer, cardiovascular disease and type 2 diabetes has historically relied on conventional screening methods such as mammograms, MRIs, ultrasounds, and blood glucose tests. While these techniques have contributed to early detection, they often come with significant limitations. These methods can be invasive, costly, and prone to issues such as patient compliance and missed diagnoses, especially in individuals at higher risk but with no apparent symptoms. Additionally, traditional risk models for cancers, such as the Gail Model for breast cancer, fail to incorporate key factors such as genetic mutations or lifestyle influences that may affect an individual’s risk profile.

[0006] Over the years, artificial intelligence (Al) has made significant inroads into the field of healthcare, particularly in cancer detection. However, the integration of Al into clinical practice has primarily focused on refining imaging techniques and improving diagnostic accuracy. Al-driven systems have the potential to analyze large datasets, such as genetic information and environmental risk factors, but this approach has not yet been fully realized in mainstream healthcare applications. Furthermore, the use of Al in screening programs has faced challenges such as the complexity of models, limited access to comprehensive data, and disparities in healthcare availability.

[0007] Despite these advancements, there remains a pressing need for a more holistic and efficient approach to early disease detection — one that can incorporate a wide range of risk factors and provide timely recommendations for at-risk individuals. These challenges in current methods highlight the potential for innovation in improving disease risk prediction and reducing healthcare burdens. SUMMARY

[0008] According to this invention, there is provided an Al-driven platform designed to enhance early detection and risk assessment for individuals at high risk of developing specific cancers (breast, prostate, colorectal, and cervical cancers), cardiovascular disease as well as Type 2 diabetes. The platform integrates multiple screening modalities into a holistic system, utilizing AI- based digital analysis to assess genetic information, risk factors, and historical health data. According to this invention, there is provided a system that leverages advanced machine learning techniques, including random forest, neural networks, and support vector machines, to process complex relationships between genetic variants and environmental risk factors, providing highly accurate risk assessments. The platform is designed to offer updated recommendations for both high-risk and low-risk individuals, guiding them on the next steps in their healthcare journey. According to this invention, there is provided a method for optimizing healthcare resources by streamlining case management, improving patient outcomes, and identifying high-risk individuals earlier. This proactive approach to healthcare enables the reduction of financial burdens on healthcare systems and frees up resources for investment in new technologies, advanced treatments, and additional staffing.

[0009] According to this invention, there is provided a scalable and efficient strategy for integrating the platform into existing screening programs and health guidelines, making it a valuable tool for healthcare providers, policymakers, and hospitals worldwide. The system is designed to address the growing global healthcare demand and to reduce costs while improving the quality of care.

[0010] BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Fig.1 : Illustrates Mammography images

[0012] Fig.2: Illustrates ROI mask of images. Right- bilateral craniocaudal (CC)(RCC) and Right- mediolateral oblique (MLO) (RMLO) views to detect tumor mass

[0013] Fig.3: Illustrates Training ECG model

[0014] Fig. 4 and Fig.5: Illustrate the representative outcomes of risk assessment in training models

[0015] DETAILED DESCRIPTION OF THE INVENTION

[0016] This platform is the world’s first integrated system that combines multiple screening modalities into a holistic framework, utilizing Al-based digital analysis to evaluate risks and identify high-risk cases for developing breast, prostate, colorectal, and cervical cancers, cardiovascular diseases, and type 2 diabetes. The platform provides updated recommendations for both high- and low-risk categories, allowing users to understand their next steps. Additionally, the system records previous history and risk assessments, enabling specialists to follow up on cases as needed.

[0017] This platform can be delivered and deployed as a web service, application, software, or any other type of service, making it adaptable to various environments and needs. The platform employs various Al and machine learning techniques, including random forest, support vector machines, neural networks, deep learning algorithms, and Natural Language Processing (NLP). These models effectively handle complex relationships between genetic variants and risk factors. For example, two CNN models were developed for colon cancer, differing only in the last fully connected (FC) layer, which determines the number of class labels in the dataset. In the colon cancer model, the final FC layer's channel size was set to two to predict colon adenocarcinoma and benign colonic tissue classes.

[0018] Federated Learning for Disease Prediction

[0019] Based on data distribution, federated learning is classified as horizontal or vertical. Horizontal federated learning is applied to homogeneous data, where nodes share the same feature space but have different samples. Vertical federated learning, on the other hand, supports heterogeneous data when nodes have overlapping samples but limited common features. This platform employs horizontal federated learning to improve prediction accuracy in cases where local training data is limited.

[0020] Federated Learning Workflow:

[0021] 1. The central server sends the global machine learning model to clients.

[0022] 2. Clients update the local model using their private datasets.

[0023] 3. Clients send their updated local models to the server.

[0024] 4. The server aggregates the local models to create a new global model.

[0025] 5. The updated global model, including metadata and parameters, is sent back to the clients.

[0026] 6. This process is repeated until a convergence threshold is met or the maximum training rounds are completed.

[0027] Federated learning ensures that training occurs on client devices, eliminating the need for direct data sharing. Instead, clients share only model updates. The FedAvg algorithm, proposed by Google, is used for aggregation, combining stochastic gradient descent updates from clients into a single global model.

[0028] This study selects Flower, a federated learning framework developed at the University of Oxford, for its ease of use and compatibility with different ML frameworks. The federated system consists of one central server and two clients, each with two ML models. The training process begins by running crc-server.py and lung-server. py, followed by crc-client.py and lung-client. py on each client. Clients train their models on local datasets, send updates to the server, and receive improved models in return.

[0029] System Architecture

[0030] The platform follows a Three-Tier Architecture, which separates business logic, data, and the user interface:

[0031] • Presentation Layer: o Purpose: Handles user interactions o Tasks: Displays information, receives user inputs, and manages user experience logic o Technologies: HTML, CSS, JavaScript

[0032] • Business Logic Layer: o Purpose: Manages business rules and data processing o Tasks: Executes business logic, validates data, and interacts with the data layer o Technologies: Python, Django

[0033] • Data Access Layer: o Purpose: Manages system data o Tasks: Stores, retrieves, updates, and deletes data while isolating access logic o Technologies: MySQL database

[0034] Software Development Lifecycle

[0035] 1. Requirements Analysis: Identifying user and stakeholder needs through interviews, literature reviews, and reverse engineering.

[0036] 2. System Design: o Defining APIs and communication protocols o MySQL database design o Django-based framework selection

[0037] 3. Software Development: o Independent module development for front-end and back-end o Version control using Git

[0038] 4. Testing: o Modular testing (unit tests) o Integration testing o User acceptance testing (UAT)

[0039] 5. Deployment & Launch: o Performance optimization o Production setup o Manual deployment to the server

[0040] 6. Maintenance & Support: o Bug fixes and performance monitoring o Periodic updates and feature enhancements

[0041] Risk Assessment Models and Technologies Used

[0042] Risk Assessment Type Technologies

[0043] Breast Cancer Python - Django - MySQL

[0044] Colorectal Cancer Python - Django - MySQL

[0045] Prostate Cancer Python - Django - MySQL

[0046] Cervical Cancer Python - Django - MySQL

[0047] Diabetes Python - Django - MySQL

[0048] Cardiovascular Disease Python - Django - MySQL

[0049] Dashboard Python - Django - MySQL - ChartJS - jscharting

[0050] Blockchain-Based EHR System

[0051] The platform integrates Electronic Health Records (EHR) with blockchain technology to enhance security, privacy, and data integrity. The system classifies users into three types: administrators, patients, and doctors, each with different roles and permissions.

[0052] Key Features for Patients:

[0053] • Add / Edit / View personal, medical, and family history

[0054] • Add / Edit / View lifestyle factors (activity levels, diet, alcohol consumption)

[0055] • Search for doctors by name, specialty, or medical center

[0056] • Grant / Revoke doctors' access to their EHR data

[0057] • View symptoms, diagnoses, and treatments history

[0058] Key Features for Doctors:

[0059] • Add / Edit / View personal information

[0060] • Accept / Reject patient access requests

[0061] • View visit history and patient list

[0062] • Add new symptoms, diagnoses, and treatments • Upload test results

[0063] Blockchain Transaction Process:

[0064] 1. Submitting Transactions: Users interact with the system via a Hyperledger Fabric SDK.

[0065] 2. Executing Smart Contracts: Predefined chaincode methods process requests (view, add, or edit data).

[0066] 3. Approving Transactions: The endorsement policy requires approval from at least one peer per organization.

[0067] 4. Updating Ledger: Approved transactions are permanently recorded on the blockchain.

[0068] AI-Based Prediction Models in Blockchain:

[0069] • If a transaction updates patient data (e.g., test results, personal history), predefined Al prediction models automatically recalculate risk scores.

[0070] • Updated risk scores are sent as notifications to users.

[0071] • Doctors can securely access patient risk assessments via smart contracts, ensuring data privacy and immutability through blockchain consensus mechanisms.

[0072] By combining federated learning, blockchain security, and Al-driven diagnostics, this platform provides a revolutionary approach to early disease detection and clinical decision-making. It also covers a wide range of screenings, including heart rate monitoring (ECG) and mammography, in addition to general screening for non-communicable diseases, Al, and image analysis in ECG and radiology.

[0073] The present platform integrates advanced machine learning algorithms, image processing techniques, and innovative data handling strategies. The novelty of the invention lies not only in the architectural design of the system but also in the unique way the algorithms have been optimized and implemented in the code. This novel approach allows for more accurate predictions, better data integration, and enhanced user interaction.

[0074] Code for Model Implementation and Training:

[0075] The following code examples represent critical components of the software system developed as part of this invention. They highlight specific aspects of the innovation, such as the application of machine learning models (CNNs, UNet), data transformation techniques, and training loops. These code snippets are intended to demonstrate the core principles of the platform's functionality. However, they constitute only a small part of the overall code base, which includes additional modules and functionalities crucial for the complete operation of the system.

[0076] Each section of the code provided corresponds to a specific innovation introduced in the invention and showcases how the platform uses these algorithms to achieve superior results. It should be noted that the overall novelty and inventive step he in the holistic integration of these components, which together form a unique and highly effective solution. o Load and show instance of data: import random import matplotlib.pyplot as pit item = random. randint(0, len(train dataset)) img, label = train_dataset[item] pit. imshow(img[0] , cmap=" gray " ) o data transformation from torchvision import transforms train transforms = transforms. Compose([ transforms. Resize((l 50, 150)), transforms. RandomHorizontalFlip(), transforms. RandomRotation(20), transforms . ToTensorQ , transforms. Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])

[0077] ]) test transforms = transforms. Compose([ transforms. Resize((l 50, 150)), transforms . ToTensorQ , transforms. Normalize([0.5, 0.5, 0.5], [0.5, 0.5, 0.5])

[0078] ]) o CNN model based on pretrained resnetl8 model import torch import torch, nn as nn from torchvision import models class CNNModel(nn.Module): def init (self, num classes, input_size=(150, 150), pretrained=True): super(CNNModel, self). init () resnetl8 = models, resnetl 8(pretrained=pretrained)

[0079] # Remove the fully connected (classifier) part of ResNetl8 selffeatures = nn.Sequential(*list(resnetl8.children())[:-l]) # Exclude the last

[0080] # Dynamically compute the input size for the linear layer dummy input = torch.zeros(l, 3, *input_size) # Adjust size to match input images features output = self. features(dummy input) flattened size = features_output.view(-l).shape[0]

[0081] # Add your custom classifier self, classifier = nn. Sequential nn.Flatten(), nn.Linear(flattened_size, 512), nn.ReLUQ, nn.Dropout(0.5), nn.Linear(512, num classes)

[0082] ) def forward(self, x): x = self.features(x) # Extract features from ResNetl8 x = self.classifier(x) # Pass through custom classifier return x o Training loop

[0083] # Training loop def train_model(model, train loader, criterion, optimizer, num_epochs=10): model. train() for epoch in range(num epochs): running loss = 0.0 correct = 0 total = 0 for images, labels in train loader: images, labels = images. to(device), labels. to(device)

[0084] # Forward pass outputs = model(images) loss = criterion(outputs, labels)

[0085] # Backward pass and optimization optimizer. zero_grad( ) loss.backward() optimizer. stepQ

[0086] # Calculate metrics running_loss += loss.item()

[0087] _, predicted = torch. max(outputs, 1) total += labels. size(O) correct += (predicted == labels). sum().item() epoch loss = running loss / len(train loader) epoch acc = correct / total print(f 'Epoch {epoch+l } / {num_epochs}, Loss: {epoch_loss:.4f}, Accuracy:

[0088] {epoch_acc:.4f}")

[0089] # Save .chpt File checkpoint = {

[0090] 'epoch': epoch,

[0091] 'model state dict' : model. state dictQ,

[0092] 'optimizer state dict' : optimizer. state dictQ, 'loss': epoch loss, torch. save(checkpoint, f 'checkpoint_epoch_{epoch} .chpt") print(f" Checkpoint saved at epoch {epoch}.") o Traning ECG model

[0093] # Train and evaluate the model train_model(model, train loader, criterion, optimizer, num_epochs=10) evaluate_model(model, test loader)

[0094] Epoch 1 / 10, Loss: 0.8289, Accuracy: 0.6702

[0095] Checkpoint saved at epoch 1.

[0096] Epoch 2 / 10, Loss: 0.4902, Accuracy: 0.8257

[0097] Checkpoint saved at epoch 2.

[0098] Epoch 3 / 10, Loss: 0.3781, Accuracy: 0.8670

[0099] Checkpoint saved at epoch 3.

[0100] Epoch 4 / 10, Loss: 0.2712, Accuracy: 0.9147

[0101] Checkpoint saved at epoch 4.

[0102] Epoch 5 / 10, Loss: 0.2727, Accuracy: 0.9090

[0103] Checkpoint saved at epoch 5.

[0104] Epoch 6 / 10, Loss: 0.2405, Accuracy: 0.9282

[0105] Checkpoint saved at epoch 6.

[0106] Epoch 7 / 10, Loss: 0.1853, Accuracy: 0.9342

[0107] Checkpoint saved at epoch 7.

[0108] Epoch 8 / 10, Loss: 0.2146, Accuracy: 0.9322

[0109] Checkpoint saved at epoch 8.

[0110] Epoch 9 / 10, Loss: 0.1878, Accuracy: 0.9335

[0111] Checkpoint saved at epoch 9.

[0112] Epoch 10 / 10, Loss: 0.1397, Accuracy: 0.9517 Checkpoint saved at epoch 10.

[0113] Test Accuracy: 97.09% o Data Transformation train transform = transforms. Compose([ transforms. Resize((224, 224)), transforms . ToTensorQ , transforms. Lambda(lambda x: (x - x.minQ) / (x.maxQ - x.min())), transforms. Lambda(lambda x: x.repeat(3, 1, 1))

[0114] ]) target transform = transforms. Compose([ transforms. Resize((224, 224)), transforms. ToTensorQ

[0115] ]) o Dataset Definition

[0116] # Dataset Preparation class Mamography Dataset(Dataset): def init (self, image_dir, mask_dir, transform=None): self.image_dir = image_dir self.mask dir = mask dir self.image filenames = os.listdir(image dir) self.mask filenames = os.listdir(mask dir) self, transform = transform def len (self): return len(self.image filenames) def getitem (self, idx): image_path = os.path.join(self.image_dir, self.image_filenames[idx]) mask_path = os.path.join(self.mask_dir, self.mask_filenames[idx]) image = Image.open(image_path).convert("L") # Grayscale mask = Image. open(mask_path).convert("L") # Grayscale if self.transform: image = self.transform(image) mask = self.transform(mask) return image, mask o Unet model for segmentation class UNet(nn.Module): def init (self): super(UNet, self). init ()

[0117] # Encoder selfencl = self.double_conv(l, 64) self.enc2 = self.down(64, 128) self.enc3 = self.down(128, 256) self.enc4 = self.down(256, 512)

[0118] # Bottleneck self, middle = self.double_conv(512, 1024)

[0119] # Decoder selfdecl = self. up( 1024, 512) self.dec2 = self.up(512, 256) self.dec3 = self.up(256, 128) self.dec4 = self.up(128, 64)

[0120] # Final Output self.final conv = nn.Conv2d(64, 1, kernel_size=l) def double_conv(self, in channels, out channels):

[0121] > Two convolution layers with ReLU activation return nn. Sequential nn.Conv2d(in_channels, out channels, kernel_size=3, padding=l), nn. ReLU (inplace=True), nn.Conv2d(out_channels, out channels, kernel_size=3, padding=l), nn. ReLU (inplace=True) ) o Unet model for segmentation - continue def down(self, in channels, out channels):

[0122] > Down-sampling using MaxPool and Double Convolution return nn. Sequential nn.MaxPool2d(2), self. double_conv(in_channels, out channels)

[0123] ) def up(self, in channels, out channels):

[0124] > Up-sampling using Transposed Convolution and Double Convolution return nn. Sequential nn.ConvTranspose2d(in_channels, out channels, kernel_size=2, stride=2), self.double_conv(in_channels + out channels, out channels) # Fixing the normalization ) def forward(self, x):

[0125] # Encoder xl = self, encl (x) x2 = self.enc2(xl) x3 = self.enc3(x2) x4 = self.enc4(x3)

[0126] # Bottleneck middle = self.middle(x4)

[0127] # Decoder with Skip Connections x = self.decl(torch.cat([middle, x4], dim=l)) x = self.dec2(torch.cat([x, x3], dim=l)) x = self.dec3(torch.cat([x, x2], dim=l)) x = self.dec4(torch.cat([x, xl], dim=l)) return torch. sigmoid(self. final conv(x))

Claims

AMENDED CLAIMS received by the International Bureau on 13 August 2025 (13.08.2025)ClaimsWhat is claimed is:

1. A comprehensive Al-based health screening system designed to identify individuals at risk of non-communicable diseases, including cancers, cardiovascular diseases, and type 2 diabetes, comprising: o A deep learning engine utilizing machine learning models, such as convolutional neural networks (CNN), support vector machines, random forest classifiers, and other relevant algorithms for analyzing medical data and identifying risk factors; o A federated learning framework that allows decentralized training of models on distributed datasets, employing horizontal federated learning to aggregate model updates from multiple client nodes while maintaining data privacy; o A blockchain-based Electronic Health Record (EHR) system ensuring secure storage, access, and management of patient data, including personal, medical, lifestyle, and genetic information, while guaranteeing data immutability and confidentiality; o A recommendation engine capable of generating personalized health assessments and follow-up actions based on the output of the risk prediction models, guiding individuals on appropriate health management steps. o an Al-based genetic counselling and pedigree analysis module for evaluating medical and genetic information2. The system of claim 1, wherein the risk assessment engine uses a combination of Al-based algorithms to evaluate the relationships between genetic, environmental, and medical factors in determining the likelihood of developing various non-communicable diseases, including cancers, cardiovascular diseases, and type 2 diabetes.

3. The system of claim 1, wherein the federated learning framework enables multiple decentralized healthcare entities, such as hospitals and clinics, to train machine learning models on local datasets, sharing only model updates to improve prediction accuracy while ensuring data privacy.

4. The system of claim 1, wherein the genetic pedigree system is configured to utilize genetic counselling technology for evaluating a genetic pedigree of a patient, and wherein smart contracts are implemented to control data processing and improve prediction accuracy, disease screenings, and healthcare consultations.

5. The system of claim 1 , wherein a limitation of current screening platforms for breast cancer and cardiovascular diseases is overcome by analyzing ECG and mammography images to improve the accuracy of prediction, disease screenings, and healthcare consultations.

6. The system of claim 1, wherein the blockchain- based EHR system utilizes blockchain technology to securely manage patient health data, with the implementation of smartcontracts to control data access, permission management, and updates, ensuring compliance with privacy and security requirements.

7. The system of claim 1, wherein the recommendation engine generates personalized recommendations for risk mitigation, including but not limited to lifestyle changes, followup screenings, and healthcare consultations, based on individual risk assessments.

8. The system of claim 1, wherein the platform integrates predictive Al models for screening various non-communicable diseases, including cancers, cardiovascular diseases, and type 2 diabetes, and incorporates the analysis of medical imaging such as ECG and radiological images as part of the overall health risk assessment.

9. The federated learning framework of claim 3, wherein the server aggregates model updates using the FedAvg algorithm to create an updated global model that is sent back to client nodes for further training.

10. The blockchain-based EHR system of claim 6, wherein patient health data is encrypted and stored in a decentralized ledger, and patients have full control over the access and sharing of their health information with healthcare providers.

11. The system of claim 1, wherein the platform incorporates medical imaging analysis, including ECG and mammography, in addition to general health screening, to provide a comprehensive risk assessment across multiple healthcare domains.

12. A method for identifying individuals at risk of developing non-communicable diseases, comprising the steps of:• Collecting patient data, including demographic, medical, lifestyle, and genetic information;• Analyzing the collected data using the Al-based risk assessment engine to calculate a risk score;• Using federated learning to train machine learning models across multiple healthcare entities, sharing only model updates;• Storing the patient data securely in the blockchain-based EHR system to ensure immutability and privacy;• Providing personalized recommendations for health management based on the calculated risk score.

13. The method of claim 10, wherein the blockchain-based EHR system allows for continuous updates to patient data and risk assessments, enabling real-time monitoring of patient health and timely modifications to care plans.

14. The system of claim 1, wherein the platform is designed for integration with existing healthcare IT systems, including Electronic Medical Records (EMR) and hospital information systems (HIS), to ensure broad compatibility and facilitate adoption by healthcare providers.[0001]Statement under Article 19(1)[0002]The amendments to the claims have been made to clarify and strengthen the scope of the invention in light of the International Search Report. Specifically, claims have been revised to:[0003]Introduce an Al-based genetic counselling and pedigree analysis module, emphasizing the evaluation of medical and genetic information.[0004]Specify the use of smart contracts within the genetic pedigree system to enhance prediction accuracy, disease screenings, and healthcare consultations.[0005]Highlight the inclusion of ECG and mammography image analysis to overcome limitations of current screening platforms for breast cancer and cardiovascular diseases.[0006]Add method steps and integration with existing healthcare IT systems to reflect the practical implementation of the platform.[0007]All amendments are fully supported by the disclosure as originally filed. No new matter has been introduced. The changes serve to clarify the claims, improve precision, and ensure that the inventive features of the platform are fully captured. These amendments do not alter the description or drawings in a way that would require substantive modification, and are intended solely to strengthen the claims for international examination purposes.

Images