Autonomous Physiotherapy Diagnosis System for Inconsistently labeled Dataset

Abstract

In this paper, we propose a novel deep learning framework for autonomous clinical diagnosis by dealing with the practical challenges in training with poor clinical dataset. The small dataset size, partially labeled data, and inconsistent labels between annotators with varying expertise make it hard to train the model to learn the effective diagnosis method in frequently and drastically updated clinical dataset. Motivated by such difficulties, the proposed framework introduces the weighted combination of inconsistent labels by taking into account the levels of annotators' expertise and adapt meta-learning approach to obtain generalized model parameters for the quick adaptation to a new task. The performance of the proposed framework is evaluated with posterior pelvic tilt detection in a squat motion, which is one of the representative rehabilitation activities. Our experimental results show that the proposed approach has a strong generalization ability and outperforms the conventional learning-based approaches, including transfer learning, in terms of the convergence speed and the converged mean squared error.