ML models can detect ALS faster, new research shows

Convolutional neural networks predict predict the best, study shows

Scientists have developed deep learning models capable of predicting the progression of amyotrophic lateral sclerosis (ALS), or Lou Gehrig’s disease.

Based in Italy, the researcher’s work found that convolutional neural networks are better for early diagnosis of ALS. The earlier the detection, the better the patient outcome as ALS can be difficult to detect with the sufferer’s muscles deteriorating over time

Findings published in Scientific Reports used neural networks to mimic how nerve cells in the brain process information. Three types of neural networks were evaluated: feed-forward, recurrent and convolutional.

The scientists built the models using the Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database – one of the largest public repositories of longitudinal data from more than 10,000 patients.

More than 250 clinical and demographic variables were factored into the neural networks to predict patient outcomes.

The neural network model was compared to the results from two established machine learning algorithms: the Bayesian Additive Regression Trees (BART) and Random Forest Regressor (RF).

Using the ALSFRS standard to determine disease progression, the researchers found that neural network models have a smaller range of error but were less accurate compared to BART. The convolutional neural network model showed the best results.

All the models demonstrated that early diagnosis was the best predictor of patient outcomes. Furthermore, the models could categorize patients into “medium-slow progressors” and “fast progressors.”

The “medium-slow progressors” showed better patient outcomes but the models couldn’t accurately predict survival outcomes.

“While deep learning [neural network] models performed comparably to state-of-the-art models, they did not provide a decisive advantage,” the researchers said.