Identification of radiographic characteristics associated with pain in hallux valgus patients: A preliminary machine learning study

ChenGuang Wang,Chao Li,Rui Zhang,ZhiJun Li,HuaFeng Zhang,Yuan Zhang,Shen Liu,XiaoYue Chi,Rui Zhao

DOI: https://doi.org/10.3389/fpubh.2022.943026

IF: 5.2

2022-08-11

Frontiers in Public Health

Abstract:Objective: To investigate the association between the structural deformity and foot pain in hallux valgus (HV) patients using a multi-variate pattern analysis (MVPA) approach. Methods: Plain radiographic metrics were calculated from 36 painful and 36 pain-free HV feet. In analysis 1, univariate analyses were performed to investigate the clinical and radiographic differences between painful and pain-free HV. In analysis 2, we investigated the pattern differences for radiographic metrics between these two groups using a MVPA approach utilizing a support vector machine. In analysis 3, sequential backward selection and exhaustive search were performed as a feature-selection procedure to identify an optimal feature subtype. In analysis 4, hierarchical clustering analysis was used to identify the optimal radiographic HV subtype associated with pain in HV. Results: We found that: (1) relative to feet with pain-free HV, the painful ones exhibited a higher hallux valgus angle, i.e., the magnitude of distal metatarsal and phalangeal deviation; (2) painful HV could be accurately differentiated from pain-free HV via MVPA. Using sequential backward selection and exhaustive search, a 5-feature subset was identified with optimal performance for classifying HV as either painful or pain-free; and (3) by applying hierarchical clustering analysis, a radiographic subtype with an 80% pain incidence was identified. Conclusion: The pain in HV is multifactorial and associated with a radiographic pattern measured by various angles on plain radiographs. The combination of hallux valgus angle, inter-phalangeal angle, distal metatarsal articular angle, metatarsal cuneiform angle and metatarsal protrusion distance showed the optimal classification performance between painful and pain-free HV.

public, environmental & occupational health