Investigators report in The American Journal of Pathology that a deep learning model trained on histological pictures of surgical tissues correctly categorized patients with and without Crohn's disease recurrence.
More than 500,000 people in the United States have Crohn's disease, according to studies. Crohn's disease is an inflammatory bowel illness that causes damage to the lining of the digestive tract. It can induce inflammation in the digestive tract, which can lead to stomach discomfort, severe diarrhea, tiredness, weight loss, and starvation.
Many patients with Crohn's disease require surgery to address their condition. Recurrence is frequent even after a successful procedure. Researchers now claim that their AI technique is extremely accurate in predicting Crohn's disease recurrence after surgery. It also found a relationship between recurrence and subserosal adipose cell histology and mast cell infiltration.
Researchers established a model that, by examining histological images, predicts the postoperative recurrence of Crohn's disease with high accuracy using an artificial intelligence (AI) tool that replicates how people view and is trained to detect and categorize photos. When comparing patients with and without illness recurrence, the AI tool discovered previously undetected changes in adipose cells as well as significant variances in the degree of mast cell infiltration in the subserosa, or gut's outer lining. The findings were published in Elsevier's American Journal of Pathology.
Crohn's disease, a chronic inflammatory gastrointestinal disorder, is thought to have a 40% postoperative symptomatic recurrence rate after 10 years. Although there are scoring techniques for measuring Crohn's disease activity and postoperative recurrence, there is no scoring system for predicting whether Crohn's disease will return.
“Most of the analysis of histopathological images using AI in the past have targeted malignant tumors,” Takahiro Matsui, MD, Ph.D., and Eiichi Morii, MD, Ph.D., both of the Department of Pathology at Osaka University Graduate School of Medicine in Osaka, Japan, discussed the study's findings. “We aimed to obtain clinically useful information for a wider variety of diseases by analyzing histopathology images using AI. We focused on Crohn’s disease, in which postoperative recurrence is a clinical problem.”
Between January 2007 and July 2018, the researchers studied 68 Crohn's disease patients who had their bowels removed. They were split into two groups based on whether or not they experienced a recurrence of illness following surgery within two years. Each group was split into two subgroups: one for AI model training and the other for model validation. For training, whole slide images of surgical specimens were cropped into tile images, annotated for the presence or absence of postsurgical recurrence, and then processed using EfficientNet-b5, a commercially available AI model designed to do image classification. When the deep learning model was evaluated using unlabeled pictures, the results showed that the unlabeled images were correctly identified according to the presence or lack of illness incidence.
The machine learning method was then used to build prediction heat maps to pinpoint locations and histological traits from which it could reliably predict recurrence. The images showed all of the layers of the gut wall. The machine learning method properly identified the subserosal adipose tissue layer, as seen by the heatmaps. In certain areas, such as the mucosal and appropriate muscle layers, the model proved less accurate. The non-recurrence and recurrence test datasets were used to choose images with the most accurate predictions. Adipose tissue was found in all of the images with the best predictive results.
The researchers speculated that subserosal adipose cell morphologies differed between the recurrence and non-recurrence groups because the machine learning algorithm made good predictions from photos of subserosal tissue. The recurrence group's adipose cells were much smaller, flatter, and had smaller center-to-center cell distance values than the nonrecurrence group's.