Local/Global explainability empowered expert-involved frameworks for essential tremor action recognition

Zhang，Lei1; Zhu，Yanjin1; Ni，Qin2; Zheng，Xiaochen3; Gao，Zhenyu1; Zhao，Qing4,5

2024-09-01

10.1016/j.bspc.2024.106457

Biomedical Signal Processing and Control   影响因子和分区

1746-8094

1746-8108

The interpretability of machine learning (ML) and neural networks has been attached great importance in medical-related applications. Considering random forest (RF) algorithm with self-explanatory function then integrating the concept of human–machine hybrid system, this paper proposes a expert-involved explainable artificial intelligence (XAI) framework for essential tremor (ET) action recognition, which assists to reveal the black-box model globally and locally, and to provide decision-making suggestions for medical practitioners. In addition, the human empirical judgment of clinical diagnosis is integrated into the action recognition process, which contributes to the model performance improvement and forms a dynamic and feedback loop. This paper collecting sensing data of six actions from 25 participants (20 ET patients and 5 healthy subjects) via smart phone monitoring system. Standardization, feature extraction and data segmentation are used to select the key features. The RF is used to train and test datasets with three different dimensions, and the six-dimensional dataset is selected for interpretable analysis, in which the expert rating data improve the recognition performance. Then, the feature importance, permutation importance, and Shapley additive exPlanations (SHAP) and local interpretable model-agnostic explanations (LIME) methods are applied from abstract to concrete for model interpretability. The model-agnostic explainable approach are relatively flexible and universal, which can visualize the features that have positive or negative impacts on the action recognition process respectively, and obtain the specific abnormal value for further diagnosis assistance. The proposed XAI framework is expected to realize the analysis of a single abnormal sample and provide more acceptable and reliable empirical suggestions for early diagnosis and healthcare.

Essential tremor recognition Explainable artificial intelligence Human–machine hybrid system Model-agnostic

SCI ; EI

英语

其他

20242116134084

Decision making ; Diagnosis ; Feature extraction ; Forestry ; Lime ; Smartphones ; Statistical tests

Medicine and Pharmacology:461.6 ; Telephone Systems and Equipment:718.1 ; Inorganic Compounds:804.2 ; Agricultural Equipment and Methods; Vegetation and Pest Control:821 ; Management:912.2 ; Mathematics:921 ; Mathematical Statistics:922.2

2-s2.0-85193738938

Scopus

1.College of Information Science and Technology,Donghua University,Shanghai,201620,China
2.Key Laboratory of Multilingual Education with AI,Shanghai International Studies University,Shanghai,201620,China
3.School of System Design and Intelligent Manufacturing,Southern University of Science and Technology,Shenzhen,518055,China
4.School of Medicine,Tongji University,Shanghai,201619,China
5.Department of Integrative Medicine for Movement Disorders,Yangzhi Rehabilitation Hospital,Shanghai,201619,China

Zhang，Lei,Zhu，Yanjin,Ni，Qin,et al. Local/Global explainability empowered expert-involved frameworks for essential tremor action recognition[J]. Biomedical Signal Processing and Control,2024,95.

Zhang，Lei,Zhu，Yanjin,Ni，Qin,Zheng，Xiaochen,Gao，Zhenyu,&Zhao，Qing.(2024).Local/Global explainability empowered expert-involved frameworks for essential tremor action recognition.Biomedical Signal Processing and Control,95.

Zhang，Lei,et al."Local/Global explainability empowered expert-involved frameworks for essential tremor action recognition".Biomedical Signal Processing and Control 95(2024).