Interactive hybrid approach to combine machine and human intelligence for personalized rehabilitation assessment

Abstract

Automated assessment of rehabilitation exercises using machine learning has a potential to improve current rehabilitation practices. However, it is challenging to completely replicate therapist's decision making on the assessment of patients with various physical conditions. This paper describes an interactive machine learning approach that iteratively integrates a data-driven model with expert's knowledge to assess the quality of rehabilitation exercises. Among a large set of kinematic features of the exercise motions, our approach identifies the most salient features for assessment using reinforcement learning and generates a user-specific analysis to elicit feature relevance from a therapist for personalized rehabilitation assessment. While accommodating therapist's feedback on feature relevance, our approach can tune a generic assessment model into a personalized model. Specifically, our approach improves performance to predict assessment from 0.8279 to 0.9116 average F1-scores of three upper-limb rehabilitation exercises (< 0.01). Our work demonstrates that machine learning models with feature selection can generate kinematic feature-based analysis as explanations on predictions of a model to elicit expert's knowledge of assessment, and how machine learning models can augment with expert's knowledge for personalized rehabilitation assessment. CCS CONCEPTS • Human-centered computing → Interactive systems and tools; • Applied computing → Health care information systems; • Theory of computation → Sequential decision making.

An Exploratory Study on Techniques for Quantitative Assessment of Stroke Rehabilitation Exercises

“…This implies the necessity of generating personalized rules for patients with various physical characteristics and functional abilities. Thus, we expect if a system can present patient's feature values on rules to therapists, therapists might be able to tune these generic rules for personalized rehabilitation assessment [16].…”

“…Decision Tree, Linear Regression, Support Vector Machine). Compared to ML-NN, the HM has a potential benefit of interpreting a model by analyzing rules of the RB model and fine-tuning a model with patient-specific rules [16]. In addition, the HM achieves good agreement with therapist 1's annotation, which is equally good with therapist's agreement (TPA) between TP 1 and TP 2.…”

“…This paper presents a hybrid model that integrates a data-driven, machine learning (ML) model with a rule-based (RB) model using a weighted average ensemble technique [2,14,16] to assess the quality of motion. For the development, we conducted a semistructured interview with therapists to elicit their knowledge of assessing rehabilitation exercises into 15 rules and collected the dataset of three upper-limb exercises from 15 post-stroke and 11 healthy subjects and the annotations of the dataset from therapists.…”

Development of an intelligent controller for robot-aided assessment and treatment guidance in physical medicine and rehabilitation

“…Exercise Type and Param. Recommendation Zhang et al [12] X Passive ROM X Guidali et al [13] ROM, static torque X Hingtgen et al [14] X ROM, velocity X Natarajan [15] X ROM, velocity X Tojo et al [16] X Force X Zariffa et al [17] ROM, smoothness, grip ability X Lee et al [18] X ROM, smoothness, compensation X Zhao et al [19] X ROM X DIAGNOBOT Active and passive ROM, force/torque These values used as inputs to a fuzzy interference system to derive the overall quality of the exercise. The use of regression analysis, to obtain predictions of clinical scores using robotic measurements, has also been studied in the literature [20,21].…”

“…Here, the measurements made during the application of the exercises recommended by the doctor are evaluated, and no treatment recommendation is made by the system. Lee et al [18] describe and evaluate an interactive hybrid approach that integrates a data-driven model with expert's knowledge on kinematic features to assess the quality of motion for stroke rehabilitation.…”