Movement-Specific Analysis for FIM Score Classification Using Spatio-Temporal Deep Learning

TL;DR

This work addresses the burden of traditional FIM assessments by introducing an automated, markerless FIM score estimation framework that uses simple actions recorded via pose estimation. The method combines a spatio-temporal graph convolutional network (ST-GCN), bidirectional LSTM (BiLSTM), and an attention mechanism to learn informative motion features from 3D skeletal data, enabling estimation of FIM motor-item scores with interpretable attention patterns. On a dataset of 277 rehabilitation patients, the approach achieved balanced accuracies in the roughly 70–79% range for best-predictive actions across transfer and locomotion items, and ablation studies showed that velocity/angle features and temporal modeling consistently improve performance. The results suggest the method can serve as a low-burden screening tool for identifying patients in need of assistance, with potential extension to multi-action inputs and home healthcare settings to further enhance FIM assessment efficiency and scalability.

Abstract

The functional independence measure (FIM) is widely used to evaluate patients' physical independence in activities of daily living. However, traditional FIM assessment imposes a significant burden on both patients and healthcare professionals. To address this challenge, we propose an automated FIM score estimation method that utilizes simple exercises different from the designated FIM assessment actions. Our approach employs a deep neural network architecture integrating a spatial-temporal graph convolutional network (ST-GCN), bidirectional long short-term memory (BiLSTM), and an attention mechanism to estimate FIM motor item scores. The model effectively captures long-term temporal dependencies and identifies key body-joint contributions through learned attention weights. We evaluated our method in a study of 277 rehabilitation patients, focusing on FIM transfer and locomotion items. Our approach successfully distinguishes between completely independent patients and those requiring assistance, achieving balanced accuracies of 70.09-78.79 % across different FIM items. Additionally, our analysis reveals specific movement patterns that serve as reliable predictors for particular FIM evaluation items.

Figures (5)

• Figure 1: Overview of the proposed method. The framework integrates motion measurement, pre-processing, and deep learning-based FIM score estimation.
• Figure 2: Measurement actions used in this study. Abbreviations for each action are shown in brackets.
• Figure 3: Illustrations showing the computation methods for (a) velocity and (b) angle features.
• Figure 4: Class-wise accuracy distributions for each action, shown by FIM item: (a) Bed, Chair, Wheelchair, (b) Toilet, (c) Tub, Shower, (d) Walk/Wheelchair, and (e) Stairs.
• Figure 5: Example of attention weights in Squat action. The visualization shows the product of spatial ($\alpha_{\tau,j}$) and temporal ($\beta_\tau$) attention weights, where darker red indicates regions of greater focus.