Planter Deformity Detection Using Improved Deep Learning Model

Introduction: Flat feet and other structural abnormalities of the foot are important factors in the occurrence of musculoskeletal disorders and reduced quality of life. Given the importance of early and accurate diagnosis of these abnormalities, this article presents a new model based on deep learning that simultaneously analyzes structural, angular, and pressure problems of the foot. Method: The model proposed in this article is as follows: first, all the collected images from the feet of healthy and unhealthy individuals are read as input. Then, using accurate segmentation techniques, the arch area of the foot is extracted from the image. Subsequently, the output of this stage is given as input to the improved YOLO model with the CBAM attention mechanism so that the diagnosis process is carried out with a greater focus on critical areas. On the other hand, for a more comprehensive analysis, the angle of the feet of the individuals is also extracted from the images and examined to measure its relationship with the pressure distributed in the foot area. Results: In the experiments conducted, the proposed model achieved an accuracy of 95.14% in identifying plantar deformities. The proposed model was also compared with other competing methods. According to the studies conducted, this method performed better than other proposed methods. Conclusion: This model, relying on a simple structure and utilizing pre - processed inputs and angular data, is able to identify plantar deformities with high accuracy and reliability. This approach can be widely used in medical screenings, medical insole design, and patient treatment monitoring. Keywords: Plantar deformity, flatfoot pressure, deep learning, YOLO model, CBAM attention mechanism. Original article Journal of Health and Biomedical Informatics 2025; 12 (2): 189 -203 doi :10.34172/jhbmi.2025.2 1 Medical Informatics Research Center, Kerman University of Medical Sciences Introduction: Flatfoot and other structural deformities of the foot are major causes of musculoskeletal disorders that can significantly impair quality of life. Early detection of these abnormalities is crucial for preventing the progression of complications and selecting appropriate treatment strategies. In recent years, the application of deep learning methods in biomechanics and medical informatics has gained significant momentum, providing powerful tools for the automated and accurate analysis of medical imaging data. This study aims to develop a novel deep learning -based model for detecting foot abnormalities, addressing structural, angular, and plantar pressure aspects simultaneously . Method: Foot images from both healthy subjects and individuals with abnormalities were collected and preprocessed. Initially, the arch region of the foot was segmented using advanced segmentation techniques. The segmented images were then passed to an enhanced YOLO architecture integrated with the Convolutional Block Attention Module (CBAM), enabling the network to focus more effectively on critical regions. Additionally, angular measurements of the foot were extracted and combined with plantar pressure distribution data to achieve a more comprehensive assessment of foot abnormalities. Results: Experimental evaluations demonstrated that the proposed model achieved an accuracy of 95.14% in detecting foot abnormalities. Comparative analyses with other state -of-the -art methods revealed that the developed approach outperformed competing techniques, not only in classification accuracy but also in computational efficiency and its ability to focus on clinically relevant regions of the foot. The integration of angular and pressure -related features with segmented image data significantly enhanced the system's robustness and precision in identifying various types of abnormalities. Conclusion: The proposed model, with its relatively simple yet effective architecture, provides a reliable solution for the accurate identification of foot abnormalities. This approach can be applied in medical screening, orthopedic insole design, and patient monitoring during rehabilitation. Furthermore, given its computational efficiency and ease of deployment, the model can be integrated into clinical environments and rehabilitation centers. Overall, this research contributes to the advancement of intelligent systems in digital health and predictive medicine.