Predicting 5-Year EDSS in Multiple Sclerosis with LSTM Networks: A Deep Learning Approach to Disease Progression.

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Informatics

Background:

• Multiple Sclerosis (MS) is a chronic neurodegenerative disease impacting quality of life due to unpredictable disability.
• Accurate prediction of MS-related disability is crucial for timely therapeutic intervention and preventing irreversible neurological damage.

Purpose of the Study:

• To predict the 5th year score of the Extended Disability Status Scale (EDSS) in Multiple Sclerosis patients.
• To evaluate the efficacy of Long Short-Term Memory (LSTM) deep learning models for MS disability prediction.

Main Methods:

• Utilized demographic and clinical data from 1000 MS patients across two centers via the MSBase database.
• Employed Long Short-Term Memory (LSTM) networks, a type of Recurrent Neural Network (RNN), to analyze longitudinal patient data.
• Applied optimization techniques and feature selection to enhance LSTM model prediction accuracy, measured by Root Mean Square Error (RMSE).

Main Results:

• An initial LSTM model achieved an RMSE of 1.46.
• Hyperparameter optimization and feature selection reduced the prediction error to 1.332.
• Key predictors for 5th year EDSS included age and pyramidal, cerebellar, sensory, and bowel/bladder function variables.

Conclusions:

• LSTM deep learning models are effective for predicting EDSS scores in MS patients.
• This approach integrates static and dynamic patient data for accurate disability prediction.
• The study demonstrates minimal prediction error, advancing MS management strategies.