Predicting amyotrophic lateral sclerosis stage based on multi-parameter ultrasound: development and validation of an...

Area of Science:

• Neurology
• Medical Imaging
• Machine Learning

Background:

• Amyotrophic lateral sclerosis (ALS) lacks objective staging tools, hindering clinical management and trials.
• Current staging relies on subjective assessments; biomarkers/imaging are often invasive or impractical.
• Ultrasound, a safe and portable modality, can detect neuromuscular changes but hasn't been used for ALS staging.

Purpose of the Study:

• To develop and validate an interpretable ultrasound model for ALS staging and risk stratification.
• To provide a non-invasive, objective tool for tracking disease progression.

Main Methods:

• 300 ALS patients were classified as early-stage (1-2) or late-stage (3-4).
• Ultrasound assessed diaphragm excursion/thickness, geniohyoid shear-wave velocity, and peripheral muscle dimensions.
• Six machine learning models predicted stage using ultrasound metrics and clinical factors, evaluated by AUC, F1, and Brier scores.

Main Results:

• A random forest model achieved an AUC of 0.843, F1 score of 0.727.
• Diaphragm excursion during deep breathing (DEDB), masseter muscle thickness (MMT), and geniohyoid shear-wave velocity (GHSWVmean) were key predictors.
• Higher DEDB, MMT, and GHSWVmean indicated earlier stages; lower peripheral muscle thickness and older age suggested late-stage disease.

Conclusions:

• Multiparameter ultrasound with machine learning offers a non-invasive, bedside tool for ALS staging.
• The model provides objective disease tracking, supporting timely interventions and patient stratification.
• This feasible approach leverages accessible ultrasound technology for routine ALS care and research.