Explainable Deep Learning Framework for Classification of Total Hip Arthroplasty Gaits

Application of clinical gait analysis (CGA) in the evaluation of Hip Osteoarthritis (HOA) is a crucial step in the identification of functional abnormalities and surgical planning through total hip arthroplasty (THA). Measured kinematic parameters provide understanding of the gait patterns that can be used for accurate diagnostics and personalized rehabilitation. Streamlined approach is essential for efficient assessment due to the substantial amount of data generated. This study aims to explore the use of Deep Learning (DL) methodology in the classification of gait patterns before and after THA. All kinematic parameters of both lower limbs are aggregated, and a 2-dimensional data representation is acquired as input to the DL model. The best classification accuracy is 98.2% from the test subset. Consequently, model prediction is explained through Local Interpretable Model-Agnostic Explanation (LIME) to determine input features that best distinguish between gait classes. Features are ranked in terms of their relevancy, and only the most common discriminating features are considered for further analysis. Generally, the sagittal hip angle is the most significant kinematic parameter across classes, albeit from different gait cycle events. All healthy and 75% of the HOA gaits in the test subset exhibit discrimination on the single limb support towards the swing phase. The sagittal angles of the knee and ankle during the same gait events are also discriminatory for HOA with 72% of the gaits exhibiting such feature. Additionally, the same kinematic angles are important for THA class prior to heel rise of the stance phase with 85% of the gaits examined. With features acquired directly from the input data, explanation is more transparent promoting trust and understanding of the DL model. High accuracy scores coupled with explainability showcases clinical relevance and a potential for aiding medical experts in crucial decision making for HOA patients before and after THA.