Can Artificial Intelligence Eliminate the Need for Postdiverticulitis Colonoscopy?
Douglas K. Rex, MD, MASGE, reviewing Ziegelmayer S, et al. JAMA Netw Open 2023 Jan 27.
Colonoscopy is often recommended after an episode of acute diverticulitis, particularly complicated diverticulitis, and when no colonoscopy was performed the previous year. Colonoscopy in prior diverticulitis can be difficult, as excellent bowel preparation may be harder to achieve with an angulated and narrowed sigmoid, and barotrauma risk is increased in diseased sigmoids.
Artificial intelligence (AI) programs are increasingly used to enhance radiographic diagnosis. This study evaluated whether an AI program could improve differentiation of acute diverticulitis from colon cancer.
Cases were acquired from a prospectively curated database. More than half were excluded for imperfect CT quality. Of the 585 included cases, 318 had cancer, 435 were used for training, 90 for validation, and 60 for testing.
For the test set, the AI system had cancer sensitivity of 83.3% and specificity of 86.6%. Board-certified radiologists had a sensitivity of 85.5%, which increased to 90.0% with AI.
![d-rex](/wp-content/uploads/d-rex.jpg)
COMMENTAlthough this is an important clinical question and AI improved radiologist sensitivity for cancer, it seems unlikely this level of AI performance would substantially impact the need for postdiverticulitis colonoscopy. First, more than half the CT scans were excluded for poor quality. Second, the prevalence of cancer was much higher in the study cases than in patients presenting with clinically acute diverticulitis. Thus, cancer sensitivity might be lower in clinical practice because of so-called “suspicion bias,” in which the prevalence of disease affects interpretation of a clinical test. Finally, after the addition of AI, the board-certified radiologists still missed two of every three cancers they had missed before AI.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
CITATION(S)
Ziegelmayer S, Reischl S, Havrda H, et al. Development and validation of a deep learning algorithm to differentiate colon carcinoma from acute diverticulitis in computed tomography images. JAMA Netw Open 2023 Jan 27. (https://doi.org/10.1001/jamanetworkopen.2022.53370)