Deep learning-based automated quantification of goblet cell mucus using histological images as a predictor of clinical relapse of ulcerative colitis with endoscopic remission.

Background: Mucin depletion is one of the histological indicators of clinical relapse among patients with ulcerative colitis (UC). Mucin depletion is evaluated semiquantitatively by pathologists using histological images. Therefore, the interobserver concordance is not extremely high, and an objective evaluation method is needed. This study was conducted to demonstrate that our automated quantitative method using a deep learning-based model is useful in predicting the prognosis of patients with UC.

Methods: Deep learning-based models were trained to detect goblet cell mucus area from whole slide images of biopsy specimens. This study involved 114 patients with UC in endoscopic remission with a partial Mayo score of ≤ 1. Biopsy specimens were collected during colonoscopy, and the ratio of goblet cell mucus area to the epithelial cell and goblet cell mucus area was calculated as goblet cell ratio (GCR). The follow-up time was 12 months, and the primary outcome was the relapse rate. Clinical relapse was defined as partial Mayo score of ≥ 3.

Results: Sixteen patients (14%) experienced clinical relapse. In the relapsed group, the GCRs of specimens obtained from the cecum, ascending colon, and rectum were significantly lower than those of specimens in the relapse-free group (p = 0.010, p = 0.027, p < 0.01). In the rectum, patients with a GCR of ≤ 12% had a significantly higher relapse rate than those with a GCR of > 12% (45% [10/22] vs. 6.5% [6/92]; p < 0.01).

Conclusions: Quantifying goblet cell mucus areas using a deep learning-based model is useful in predicting the clinical relapse in patients with UC in clinical and endoscopic remission.