Scholarlink

This chapter presents an investigative report on Hip Osteoarthritis (HOA) gait kinematic parameters and its utilization in the context of a classification problem. Assessing HOA objectively is crucial in the diagnostics, treatment decisions, and rehabilitation efforts. However, there is still a lack of consensus on the relevant gait parameters that aptly describe the disease. The objective of our research is to use Deep Learning (DL) methodology to determine kinematic gait parameters with significant discriminating value, to classify HOA and healthy subjects. Based on the results from the literature, kinematic trajectories from the ankle, foot progression angle (FPA), hip, knee, and pelvis are selected, and transformed into time–frequency domain through Continuous Wavelet Transform (CWT). The resulting scalogram images of both limbs are merged through alpha blending and used as inputs to a DL model for classification. A lightweight DL model, LightGaitNet is designed based on the strengths of state-of-the-art Convolutional Neural Network (CNN) pre-trained models, namely ResNet50 and Inception architecture, to reduce computational complexity while maintaining accuracy. It was found out that the sagittal angles of ankle, knee, and hip are the most discriminating parameters with more than 91% of \({G}_{mean}\) scores. The frontal angle of the hip and the pelvis are also consistent on the different models considered. Several state-of-the-art pre-trained CNN models are also trained and compared with the proposed design. It was found that the same kinematic parameters resulted in high performance across the different models validated, thus boosting the utility on clinical settings. To improve trustworthiness of the results, a thresholding method is applied utilizing softmax layer results to remove predictions with low probability. Gait samples with low confidence are pruned from the results. It is further increasing performance to 96%, on the best performing kinematic parameters. The presented novel DL methodology can further be utilized when gaits are examined in the clinical setting. While the results agree with most discriminating features reported in the literature, it was concluded that, the sagittal angle of the ankle is also a crucial discriminating kinematic parameter that should be used in the classification.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Light Deep Learning Methodology in the Classification of Hip Osteoarthritis Gait Kinematic Data

  • 章节

  • 发表日期：

    2026年

查看全文
  • Roel Pantonial,
  • Milan Simic

摘要

This chapter presents an investigative report on Hip Osteoarthritis (HOA) gait kinematic parameters and its utilization in the context of a classification problem. Assessing HOA objectively is crucial in the diagnostics, treatment decisions, and rehabilitation efforts. However, there is still a lack of consensus on the relevant gait parameters that aptly describe the disease. The objective of our research is to use Deep Learning (DL) methodology to determine kinematic gait parameters with significant discriminating value, to classify HOA and healthy subjects. Based on the results from the literature, kinematic trajectories from the ankle, foot progression angle (FPA), hip, knee, and pelvis are selected, and transformed into time–frequency domain through Continuous Wavelet Transform (CWT). The resulting scalogram images of both limbs are merged through alpha blending and used as inputs to a DL model for classification. A lightweight DL model, LightGaitNet is designed based on the strengths of state-of-the-art Convolutional Neural Network (CNN) pre-trained models, namely ResNet50 and Inception architecture, to reduce computational complexity while maintaining accuracy. It was found out that the sagittal angles of ankle, knee, and hip are the most discriminating parameters with more than 91% of \({G}_{mean}\) scores. The frontal angle of the hip and the pelvis are also consistent on the different models considered. Several state-of-the-art pre-trained CNN models are also trained and compared with the proposed design. It was found that the same kinematic parameters resulted in high performance across the different models validated, thus boosting the utility on clinical settings. To improve trustworthiness of the results, a thresholding method is applied utilizing softmax layer results to remove predictions with low probability. Gait samples with low confidence are pruned from the results. It is further increasing performance to 96%, on the best performing kinematic parameters. The presented novel DL methodology can further be utilized when gaits are examined in the clinical setting. While the results agree with most discriminating features reported in the literature, it was concluded that, the sagittal angle of the ankle is also a crucial discriminating kinematic parameter that should be used in the classification.