Determination of Ataxia with EfficientNet Models in Person with Early MS using Plantar Pressure Distribution Signals

Abstract

Multiple Sclerosis (MS) is a central nervous system disease that causes ataxia and balance disorders. In ataxia, the first symptom is usually seen as gait disturbance. In gait ataxia, symptoms can be clinically defined by shortened stride length and irregular strides. Evaluation of gait disturbance in clinical cases is important for the detection of the first stage of ataxia. With the increasing amount of data, high-performance models can be produced, especially in the field of healthcare, with computer machine learning, deep learning and artificial intelligence methods. This study aimed to identify ataxia in individuals with Multiple Sclerosis (MS) by analysing images that encompass plantar pressure distribution signals. A total of 105 images, each containing plantar pressure distribution signals, were utilized to extract features through pre-trained EfficientNet architectures. Then the feature vectors obtained were classified by SVM, k-NN, and ANN methods. As a result of this study, the best classification performance was obtained with SVM classifier with 88.09 % Acc, 80.55 % Sen, 93.75 % Spe and 85.29 % F1 Score. The results show that the study will help the clinician in the detection of PwMS ataxia and will be a pioneer for future studies.

Keywords

• deep learning
• feature extraction
• image classification