Machine Learning Integrated Software for Prediction, Diagnosis, and Prognosis of Amyotrophic Lateral Sclerosis

Overview

Our software system is designed to assist medical personnel in early diagnosing and prognosticating of Amyotrophic Lateral Sclerosis (ALS), a neurodegenerative disease that is often misdiagnosed or overlooked. The software features five tabs, each with distinct functions: three for users and two for administrators. Tab 3 predicts the likelihood of ALS using clinical data including electromyography (EMG), blood, and urine tests, analyzed through Neural Networks and Logistic Regression. Tab 4 uses Decision Trees and K-Nearest Neighbors algorithms, leveraging the ALS Functional Rating Scale-Revised to determine patient stages. Tab 5 employs Linear Regression to estimate patient remaining time, utilizing specially refined indices. The input datasets contain basic, easily collectible, and cost-effective data, avoiding complex tests like MRI and spinal tap, which makes it particularly suitable for detecting early-stage symptoms such as muscle twitches. The data collection process involved compiling demographic features and clinical tests from a control group of 300 individuals without neurological disorders and an experimental group of over 11,700 ALS patients sourced from the PRO-ACT global database and Vietnam hospitals. Data were prepared and standardized for model input, with noise filtering, model training, and validation taking place in tab 2 to ensure optimal index compatibility and maximum accuracy. After training, tab 1 is used to identify, implement, and customize the optimal hyperparameters for each model. Through flexible adjustments in data selection, model settings, and fine-tuning, an advanced tool has been developed achieving accuracy rates of up to 82%, 91%, and 78% for ALS prediction, diagnosis, and prognosis, respectively.

Competition history

• ISEF 2024 Translational Medical Science · Entry TMED074T · Los Angeles, California, United States

Resources

• ISEF abstract page ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.