Enhancing Fall Risk Assessment in the Elderly: A Study Utilizing Transfer Learning in an Improved EfficientNet Network with the Gramian Angular Field Technique

Abstract

Recent years have seen a significant increase in the incidence of falls among the elderly, leading to accidental injuries and fatalities. This trend underscores the critical need for accurate fall risk assessment, a major concern for public health and safety. In addressing this challenge, a novel approach has been developed, leveraging a pressure sensor placed on the foot's sole to gather gait data from elderly individuals. This method provides a precise analysis of gait stability on a daily basis. The research introduced here utilizes the gramian angular summation field (GASF) technique for converting this data into two-dimensional images, which are then processed using an enhanced EfficientNet model. The innovation lies in the integration of a convolutional block attention module (CBAM) into this model, resulting in a CBAM-EfficientNet algorithm. This approach includes freezing the first four stages of the EfficientNet model, focusing training on the deeper layers that incorporate CBAM. This strategy is aimed at augmenting the network's ability to extract critical features from foot pressure data, consequently improving the accuracy of fall risk classification. Experimental evaluation of this optimized model demonstrates a classification accuracy of 98.5% and a sensitivity of 99.0%, indicating its practical efficacy and strong generalization capacity. These findings reveal that the methodology significantly enhances the classification of plantar pressure data, offering valuable support in medical diagnosis and has substantial practical implications.