Color image processing in Hirschsprung's disease diagnosis

Abstract

Hirschsprung's disease is a gut motility disorder, which can affect newborns and children, characterized by absent ganglion cells within the rectum or colorectum. Surgical intervention involves a pull-through operation, connecting ganglionic segments together after excising the aganglionic segment. Subjective examination of seromuscular biopsies and resected segment for ganglion cells by a pathologist is the standard of care for Hirschsprung's patients. This paper explores the feasibility of an objective approach using multistage color image processing to segment the muscularis propria, identify regions of interest within, separate plexuses in the regions of interest, and detect and quantify ganglion cells within individual plexuses. Results observed on one test case showed that this multistage approach was able to segment muscularis propria with results comparable to manual segmentation at 77.3% region-coincidence. Regions of interest were identified with 100% accuracy, all containing at least one plexus, with 0 false negatives. Automatic plexus segmentation had a precision of 88.5% and recall of 90.2%. Automated ganglion detection achieving a precision of 85.7% and recall of 72.0%. Preliminary results are encouraging but performance needs improvement. The main issues encountered is the varying colour contrast within an image and the use of an imperfect feature set for classification.