Hyperspectral Imaging Assessment of Systemic Sclerosis Using the Soft Abundance Score and Band Selection

Abstract

Hyperspectral imaging (HSI) is an optical remote sensing technology that has the advantages of high spatial and spectral resolution. Aside from its use in geographical research, HSI has been widely used in medical diagnosis. Systemic sclerosis (SSc) is a multiorgan autoimmune disease that leads to skin tightness, thickness, and fibrosis. Internal organ involvement and mortality in this progressive disease are strongly correlated with the extent and severity of abnormal skin thickness. Our prior study demonstrated that HSI can outperform conventional assessment tools as a diagnostic modality to evaluate the severity of skin sclerosis in SSc patients. However, the analysis algorithm has its limitations. This study aimed to investigate a novel soft abundance score for HSI. We also explored the influence of band selection on the HSI analysis of SSc patients. In total, we enrolled 30 SSc patients (male: 10; female: 20, median age±range, 49.9±17.0 years) and 24 healthy controls (male: 12; female: 12, median age±range, 37.0±11.0 years). We found that most of the spectral bands generated by different band selection methods were similar. Moreover, in the task of distinguishing SSc patients from healthy controls, the soft abundance scores calculated from these bands exhibited greater discriminative power than a spectral angle mapper (SAM), skin scores determined by clinical assessments, or skin thickness determined through ultrasonography. Our results suggest that the spectral bands selected in this study should be taken into consideration to guide future hardware improvement. The analytic algorithm can also be applied as a new clinical method.