Ulcerative colitis (UC) is a chronic inflammatory bowel disease (IBD) affecting the colon and rectum, resulting in abdominal pain, diarrhea, and rectal bleeding.
The disease is characterized by a remitting-relapsing course, where episodes of disease activity or flares punctuate periods of remission. The treatment of UC aims to extinguish inflammation to prevent complications, and histopathology is considered the most stringent method to detect the presence of inflammation and distinguish it from remission.
According to research published in Gastroenterology, grading severity remains challenging, as over 30 histological indices have been proposed, none of which are ideal, and their adoption in clinical practice remains modest. Scoring is time-consuming, requires dedicated training, and is limited by high interobserver variability. This has led to clinical trials resorting to expensive centralized readings to attempt reliable measurements.
To address this challenge, computer-aided diagnosis (CAD) systems based on artificial intelligence (AI) are increasingly being used to simplify and standardize the evaluation of medical imaging. AI models have successfully been applied in digital pathology to quantify the expression of molecular targets, such as hormone receptors and HER2 in breast cancer or protein Ki67 in carcinoid tumors, and to automate morphological analysis of nuclei and cellular features. These technologies promise to enhance assessment, simplify interpretation and resolve discrepancies between pathologists.
However, in UC pathology, only two AI models have been developed. The first focused on detecting eosinophils and their correlation with disease activity, and the second concentrated on neutrophils as hallmarks of activity. Therefore, a comprehensive study of digital pathology computerized image analysis with a new and improved model to detect UC disease activity as defined by different histologic indices was needed.
This study developed a new AI-enabled model to detect UC disease activity as defined by different histologic indices, including the PHRI, Robarts Histologic Index (RHI), and Nancy Histologic Index (NHI). The model was trained on a large dataset of digitized colon biopsy images and validated on an independent cohort. The results showed that the AI-enabled model could accurately detect UC disease activity as defined by the histologic indices, with high sensitivity and specificity. Moreover, the model could also accurately forecast disease flare-ups indicated by pre-specified clinical outcomes.
The findings of this study suggest that AI-enabled models can enhance the assessment of UC pathology, simplify interpretation, and potentially resolve discrepancies between pathologists. The development and validation of this model represent a significant step towards implementing AI-enabled tools in clinical practice, providing a valuable aid in diagnosing and managing UC.
A team of researchers has developed an advanced AI-based CAD system capable of analyzing digitized biopsies to detect ulcerative colitis (UC) disease activity, estimate endoscopic activity, and predict future clinical outcomes. The model was developed using a new scoring index, PHRI, explicitly designed to be implementable into machine learning models.
The CAD system showed solid diagnostic performance in detecting disease activity with an overall AUROC of 0.87, a sensitivity of 89%, and a specificity of 84%. The system was also able to predict the presence of endoscopic inflammation with around 80% accuracy and stratify the risk of flare based on histological data.
The researchers externally validated the cross-sectional histological assessment and longitudinal prognostic stratification of the AI, which identified patients at risk of flare similarly to human pathologists. The tool represents an important step towards standardizing histological reading in clinical trials and improving patient outcomes.
