Multimodal Classification System for EDSS Score in Multiple Sclerosis: Integrating Data Encryption and Machine Learning

This project is an academic project from the Big Data course at the Avellino Campus of the University of Salerno, developed and maintained by DomFalco and open-sourced on GitHub. Its core content is to combine multimodal machine learning models with data encryption technology to realize automatic classification of EDSS scores for Multiple Sclerosis (MS) patients, balancing model accuracy and patient privacy protection.

Keywords: Medical AI, Multiple Sclerosis, EDSS Score, Multimodal Model, Data Encryption Original Link: https://github.com/DomFalco/Classificazione-dell-EDSS-nella-sclerosi-multipla-con-modello-multimodale-e-cifratura-dei-dati

Disease and Assessment Needs

Multiple Sclerosis (MS) is a chronic autoimmune disease of the central nervous system and one of the main causes of disability in young adults. Accurate assessment of disability level is crucial for treatment, prognosis, and research.

EDSS Scoring System

The Expanded Disability Status Scale (EDSS) is a standard tool for MS disability assessment, with scores ranging from 0 (normal) to 10 (death), covering 8 functional systems such as vision and brainstem.

Limitations of Traditional Assessment

Relying on clinical examinations by neurologists, there are problems such as strong subjectivity, long time consumption, and need for professional training. An automated classification system has clinical value.

Multimodal Data Fusion

Integrate four types of patient data:

• Clinical Data: Structured text such as medical history, symptoms, and treatment records
• Imaging Data: Medical image features like MRI
• Laboratory Indicators: Blood and cerebrospinal fluid test results
• Functional Assessment: Quantitative indicators from walking tests and cognitive assessments

Data Encryption and Privacy Protection

• End-to-end encryption covers the entire process of data collection/transmission/storage/processing
• Explore homomorphic encryption to realize inference in encrypted state
• Introduce differential privacy to prevent models from memorizing sensitive information
• Strict access control to ensure authorized access

Machine Learning Model

• Feature Extraction: Transformer for text, CNN for images
• Fusion Layer: Attention mechanism/gating strategy to dynamically integrate multimodal features
• Classification Layer: Supports discrete classification or continuous regression
• Uncertainty Quantification: Output confidence for doctors to review

Data Collection and Preprocessing

Automatically collect data from systems like HIS/PACS, and complete cleaning, standardization, and missing value handling.

Automatic Classification Inference

After inputting multimodal data, feature extraction, fusion, and classification are completed in an encrypted environment, outputting EDSS scores and confidence levels.

Result Interpretation and Visualization

• Feature importance identification
• Image attention heatmap
• Similar case comparison analysis

Clinical Decision Support

Integrate clinical guidelines to provide personalized treatment recommendations and follow-up plan suggestions.

Data Scarcity

• Transfer Learning: Fine-tuning pre-trained models
• Data Augmentation: SMOTE oversampling and medical-specific strategies
• Multi-center Cooperation: Expand training samples

Modal Missing Handling

Supports flexible input and adapts to scenarios with partial modal missing.

Model Interpretability

Use attention visualization and SHAP value analysis to improve decision transparency.

Educational Significance

Provide students with full-process practice in medical AI (data engineering, machine learning, privacy protection).

Clinical Translation Potential

After clinical validation and regulatory approval, it is expected to become a clinical auxiliary tool.

Technology Promotion Value

The multimodal fusion and privacy protection scheme can be extended to intelligent assessment systems for other chronic diseases.

This project realizes automatic classification of EDSS scores for MS patients through multimodal machine learning, combining advanced encryption technology to protect privacy, reflecting the development trend of medical AI: 'technical performance + data security + ethical compliance'. With the digital transformation of healthcare, such intelligent auxiliary systems will improve the quality and efficiency of medical services.