Novel AI System Created to Classify Inflammatory Bowel Disease Neoplasia Lesions, Increased Diagnosis Accuracy

In recent years, Japan has seen a significant increase in the incidence of inflammatory bowel diseases - an intractable disease characterized by chronic inflammation of the GI tract. Chronic inflammation linked with IBD leads, more often than not, to the development of colorectal cancer.

IBD and Cancer

(Photo: Vidal Balielo Jr. from Pexels)

For patients diagnosed with visible or low-grade dysplasia, an abnormal cell growth, endoscopy resection, a technique used to remove lesions, and colonoscopy, are conventionally used. However, for patients diagnosed with a high rate of neoplasia, a total proctocolectomy, the removal of the colon and rectum, is the standard for treatment, which is highly detrimental to the patient's quality of life.

Hence, identifying the severity and grade of the neoplasia during the patient's diagnosis is essential before treatment can proceed. Unfortunately, inflammation in the patient's colorectal region makes it too challenging for endoscopists to classify the type of IBD neoplasia, leaving biopsy as the only viable option, a procedure associated with high risks often leading to inaccurate diagnoses.

ALSO READ: Novel System Improves Breast Cancer Diagnoses While Decreasing Doctor's Workload

AI System Allows Accurate Diagnosis of Type and Level of Neoplasia

Because of the challenges medical professionals face, researchers from Okayama University Graduate School of Medicine, including Hideaki Kinugasa, Doctor Shumpei Yamamoto, Professor Yoshiro Kawahara, and Professor Sakiko Hiraoka, conducted a pilot study developing an AI system that can classify IBDN lesions more accurately and less invasive published in the journal Gastroenterology and Hepatology, titled "The diagnostic ability to classify neoplasias occurring in inflammatory bowel disease by artificial intelligence and endoscopists: A pilot study."

Researchers used a conventional neural network for analyzing visual imagery, known as Efficient-Net-B3, to develop a prototype for the AI system. The team then trained the system using 862 endoscopic images of 99 IBDN lesions from patients diagnosed with IBD in two hospitals between 2003 to 2021, validating using a deep-learning framework.

Next, the team asked endoscopists with more than eight years of experience in HI endoscopy to analyze the images and classify the lesions into two types based on the need for proctocolectomy and proceeded to compare the human classification to the AI system.

As a result of data augmentation, the newly developed AI system generated roughly six million images from the original data set, which were used further to analyze the clinicopathological characteristics of patients and their lesions.

Based on the analysis by the AI system, researchers found that most patients had ulcerative colitis, a type of IBD, with more than 95% of the patients presenting pancolitis and left-sided colitis. Furthermore, the AI system displayed image-based diagnosis abilities with 64.5% sensitivity and 89.5% specificity, and 80.6% accuracy. The correct diagnosis rate of the AI system developed was 79 compared to the endoscopist's results of 77.8

Assistant Professor Kinugasa explains that the AI system successfully determines the degree of malignancy of IBD tumors and is valuable enough to contribute to clinical practices in the coming years, reports NewsMedical.