💻 computer science

Heart disease detection with trust-based authentication framework in Edge-cloud-IoMT using Adaptive Aggregation Multi-scale Fusion Network

This paper proposes a trust-based authentication framework integrated with an Adaptive Aggregation Multi-scale Fusion Network (AAMF-Net) within an Edge-cloud-IoMT architecture to effectively detect heart disease, achieving 96.37% accuracy by addressing edge resource constraints through advanced data preprocessing, feature selection, and a hybrid deep learning model combining self-attention and multi-scale fusion.

Original authors: Neha Sambhaji Suryavanshi, Suresha D., Amolkumar N. Jadhav, Anushka D. Kadage, Mukesh Kumar Tripathi, Bibhuprasad Sahu

Published 2026-10-13

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Neha Sambhaji Suryavanshi, Suresha D., Amolkumar N. Jadhav, Anushka D. Kadage, Mukesh Kumar Tripathi, Bibhuprasad Sahu

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Heart disease remains a leading cause of death worldwide, and catching it early is often the difference between life and death. For decades, doctors have relied on patient history and physical exams to spot warning signs, but the sheer volume of data generated by modern medical devices offers a new path forward. This is where the Internet of Medical Things, or IoMT, comes in. Imagine a network of smart sensors and wearable devices that continuously monitor a patient's heart, sending a steady stream of information to a central system. The challenge has always been how to process this flood of data quickly and securely without overwhelming the systems that store it. Traditional methods often struggle with the sheer scale of information or fail to protect sensitive patient records from unauthorized access. Furthermore, many existing computer programs used to analyze this data are either too slow to be useful in an emergency or lack the precision to distinguish between a healthy heart and one in trouble.

A team of researchers has developed a new system designed to solve these specific problems by combining three layers of technology: the devices themselves, a local processing layer called "edge" computing, and a central "cloud" server. In their recent work, they introduced a sophisticated new model called AAMF-Net, which acts as a highly trained digital diagnostician. The system is built to work within a secure environment where patient data is protected by a trust-based authentication framework. Before any medical data is analyzed, the system verifies the identity of the device and the user, ensuring that only authorized information enters the network. This security layer uses a complex process to generate unique digital keys, effectively locking the data so that only the intended medical systems can read it. Once the data is secure, it moves to the heart of the operation: the detection of heart disease.

The researchers did not simply feed raw numbers into a computer. They first cleaned the data, removing errors and filling in missing pieces to create a clear picture of the patient's health. They then balanced the importance of different health factors, ensuring that no single symptom dominated the analysis. To handle the fact that medical records often have more examples of healthy hearts than diseased ones, the system generated additional, realistic synthetic examples of heart disease to teach the model what to look for. This preparation allowed the core of their system, the AAMF-Net, to learn effectively. This model is a hybrid of two advanced techniques. One part focuses on understanding the deep, long-term patterns in the data, while the other part pays close attention to the fine details and edges of the information, much like how a human expert might look at both the big picture and the specific symptoms. By merging these two perspectives, the system creates a comprehensive view of the patient's condition.

When the researchers tested this new approach, the results were striking. They compared their system against several other established methods using real-world medical datasets containing thousands of patient records. The new model proved to be more accurate and reliable than the previous techniques. In their tests, the system correctly identified heart disease cases with an accuracy of 96.37% and successfully detected the presence of the disease in 97.13% of the cases where it was actually present. These numbers represent a significant improvement over older models, which often struggled with noisy data or missed subtle signs of illness. The researchers also used statistical tools to confirm that these results were not just a lucky guess but a consistent performance. The system was able to explain its decisions by highlighting which specific health factors, such as age or cholesterol levels, were most important in its diagnosis, making it a trustworthy tool for medical professionals.

The study demonstrates that by carefully securing data and using a dual-layered approach to analyze it, it is possible to build a system that is both fast and incredibly precise. The researchers simulated the entire process, from the moment a device collects data to the final diagnosis, showing that this method works well even when the data is imperfect or incomplete. While the work is currently a simulation and not yet a clinical trial in a hospital, the findings suggest a clear path forward for how technology can assist doctors in saving lives. The system does not replace the human physician but offers a powerful, secure, and highly accurate assistant that can help identify heart disease earlier than ever before.

Technical Summary: Heart Disease Detection with Trust-Based Authentication in Edge-Cloud-IoMT using AAMF-Net

1. Problem Statement

Heart disease remains a leading cause of global mortality, necessitating early identification and continuous monitoring to improve patient outcomes. While the Internet of Medical Things (IoMT) combined with edge and cloud computing offers promising avenues for remote diagnosis, conventional approaches face significant hurdles:

Resource Constraints: Edge devices often struggle with scalability and computational limitations.
Security and Privacy: Clinical data is highly sensitive. Existing frameworks often lack robust trust-based authentication, leaving systems vulnerable to unauthorized access and data leakage.
Data Quality Issues: Conventional models frequently suffer from incomplete or noisy data, class imbalance, and a lack of interpretability, leading to reduced diagnostic accuracy.
Latency and Efficiency: Relying solely on cloud processing introduces latency, while edge-only solutions may lack the computational power for complex deep learning tasks.

2. Methodology

The proposed research introduces a comprehensive framework for heart disease detection within an Edge-Cloud-IoMT ecosystem, centered on a novel Adaptive Aggregation Multi-scale Fusion Network (AAMF-Net). The methodology is structured into three primary phases:

A. Trust-Based Authentication and Key Generation

Before data processing, a secure environment is established involving Cloud, Edge, and IoMT entities:

Simulation & Registration: Entities (Cloud, Edge, IoT) undergo a registration process involving ID/password storage and One-Time Password (OTP) verification.
Key Generation: A Deep High-order Attention Neural Network (DHA-Net) is utilized to generate private keys, capturing complex multi-level feature interactions to enhance security.
Authentication Protocol: The system employs a multi-step authentication process (Initialization, Registration, Key Generation, Authentication, Authorization, Access Control, and Data Protection).
Security Operations: The framework integrates DES encryption, hashing functions, XOR operations, and OTPs. Trust is established through direct, indirect, mediation, and attestation trust mechanisms.

B. Data Pre-processing and Augmentation

Secured data from the cloud undergoes rigorous preparation to ensure model robustness:

Pre-processing:
- Outlier Removal: Utilizing the Z-score technique to eliminate anomalies in continuous features.
- Missing Value Handling: Applying median imputation for numerical features and mode imputation for categorical features.
- Categorical Encoding: Converting non-numeric variables using Learned Embedding (for high-cardinality features) and One-Hot Encoding (for features with ≤5 distinct values).
Feature Equalization: Variable Stability Scaling (VSS) is applied to standardize feature contributions by adjusting weights based on consistency across samples, incorporating the coefficient of variation.
Dimensionality Reduction: Supervised Infinite Feature Selection (Inf-FSs) is employed to rank and select features based on graph-based filtering, considering normalized standard deviation, Fisher criteria, and mutual information.
Synthetic Data Generation: To address class imbalance, Finite Elements-Synthetic Minority Oversampling Technique (FE-SMOTE) generates synthetic samples by constructing simplexes within the feature space, ensuring new samples occupy boundary and interior areas of triangulation.

C. The AAMF-Net Architecture

The core detection model, AAMF-Net, is a hybrid architecture designed to fuse multi-scale features and edge information. It integrates two sub-networks:

Adaptive Aggregation with Self-Attention Network (AASAN):
- Utilizes a ResNet backbone with Self-Attention with Relative Position Encoding (SA-RPE) to capture long-term dependencies.
- Employs a Global Branch to generate attention maps and a Local Branch to extract focal features, suppressing background noise.
- Uses an Adaptive Aggregation Feature (AAF) component to semantically merge unreliable features.
Edge-guided Multi-scale Attention Fusion Network (EdgeMAF-Net):
- Designed to capture complex cardiac irregularities using multi-scale features.
- Incorporates a Cross-stage Partial Fusion Module (CSPFM) and a Multi-scale Edge Attention Calibration Module (MSEACM).
- CSPFM includes a High-Frequency Attenuation (HFA) layer using Fourier transforms to suppress noise while preserving edges.
- MSEACM utilizes Adaptive Gating Fusion (AGFB), Feature Enhancement Attention (FEAB), and Multi-scale Edge Enhancement (MSEEB) to refine feature maps.
Fusion Mechanism:
- The outputs of AASAN ( $A_h$ ) and EdgeMAF-Net ( $G_h$ ) are combined using a Taylor concept approximation.
- The final detection output ( $R_h$ ) is derived via the relationship $R_h = 2A_h - G_h$ , effectively fusing the interpretability of AASAN with the edge-aware precision of EdgeMAF-Net.
- The model is optimized using the Adam optimizer and a Focal Loss function to handle class imbalance.

3. Key Contributions

AAMF-Net Architecture: The proposal of a novel hybrid network combining AASAN and EdgeMAF-Net to detect heart disease, leveraging both self-attention mechanisms and edge-guided multi-scale fusion.
Integrated Security Framework: Implementation of a trust-based authentication system using DHA-Net for key generation, ensuring data integrity and privacy across the Edge-Cloud-IoMT spectrum.
Advanced Data Handling: A pipeline featuring VSS for feature equalization, Inf-FSs for supervised feature selection, and FE-SMOTE for generating high-quality synthetic data to resolve class imbalance.
Interpretability: The inclusion of SHAP (SHapley Additive exPlanations) analysis to validate the clinical relevance of selected features (e.g., Oldpeak, thalch, serum cholesterol).

4. Experimental Results

The model was evaluated on two datasets: UCI Heart Disease Data and Cardiovascular_Disease_Dataset.

Performance Metrics:
- Accuracy: AAMF-Net achieved 96.37% accuracy on the Cardiovascular_Disease_Dataset and 96.25% on the UCI dataset.
- Detection Rate: The model attained a detection rate of 97.13% (Cardiovascular) and 97.02% (UCI).
Comparative Analysis: AAMF-Net outperformed existing methods including FRIEND, Explainable-IoT framework, CNN, and Hybrid mutual authentication approaches. For instance, on the Cardiovascular dataset, AAMF-Net improved accuracy by approximately 6.79% over the FRIEND model and 2.03% over the Hybrid mutual authentication approach.
Ablation Study: The full AAMF-Net model consistently outperformed its individual components (AASAN and EdgeMAF-Net alone), confirming the efficacy of the fusion strategy.
Scalability and Calibration: The model demonstrated stable performance across varying feature dimensions and produced reliable probability estimates, as evidenced by calibration curves closely following the ideal 45° line.
Statistical Significance: T-tests confirmed that the improvements in accuracy and detection rate were statistically significant compared to baseline models.

5. Significance and Claims

The paper claims that the proposed AAMF-Net framework offers a robust, secure, and accurate solution for heart disease detection in Edge-Cloud-IoMT environments. By integrating a trust-based authentication mechanism with a sophisticated deep learning architecture, the research addresses critical gaps in scalability, data security, and diagnostic precision.

The authors emphasize that the system enables:

Early Detection: Facilitating timely intervention through continuous monitoring.
Enhanced Security: Protecting sensitive patient data through advanced key generation and trust mechanisms.
Clinical Reliability: Providing interpretable predictions that align with clinically significant features, thereby supporting healthcare providers in decision-making.

The study concludes that while big data analytics and further risk factor prediction are potential future directions, the current implementation successfully demonstrates a significant advancement over conventional heart disease detection methods.