Editors: Liron Pantanowitz, MD, PhD, MHA, chair of the Department of Pathology and professor of pathology, University of Pittsburgh Medical Center, and Matthew G. Hanna, MD, vice chair of pathology informatics and associate professor, Department of Pathology, University of Pittsburgh Medical Center.
Use of AI to detect ganglion cells in diagnosis of Hirschsprung disease
April 2025—Hirschsprung disease is characterized by the absence of ganglion cells in the intestinal wall. Determining whether ganglion cells are present in an effort to identify Hirschsprung disease is a cumbersome task for pathologists that may require frozen section analysis; histopathologic assessment of a rectal biopsy specimen (the gold standard); use of special stains, such as AChE; IHC analysis of calretinin or S100; or molecular and genetic testing. To assist pathologists with evaluating challenging frozen sections, the authors developed an artificial intelligence (AI) solution designed to enhance the detection of ganglion cells during intraoperative consultation. The AI model was trained using a mixed data set that combined 366 frozen section and 302 formalin-fixed, paraffin-embedded H&E-stained slides procured from 164 patients across three medical centers in Turkey. After scanning the slides, pathologists helped train the deep learning model by annotating ganglion cells present in the whole slide images (WSI). Large WSI files were also segmented and split into smaller image patches. An algorithm for generating heat maps to localize ganglion cells was employed. By directing pathologists’ attention to regions containing such cells, the model minimized the need for pathologists to review the entire WSI. Frozen section diagnoses were confirmed by studying the corresponding permanent sections. The AI-based approach achieved accuracy of 91.3, 92.8, and 90.1 percent in detecting ganglion cells within WSI across the three medical centers. In a multi-reader study, 10 pathologists assessed 50 frozen section WSI—25 with and 25 without ganglion cells. The participating pathologists initially examined these digital slides without using AI. After a washout period of at least two weeks, they were asked to use AI to re-evaluate the same digital slides, along with four image patches that displayed the highest probability of containing ganglion cells. In this reader study, the pathologists’ diagnostic accuracy, aided by the AI model’s heat maps and patches, increased from 77 to 85.8 percent. Of note, the time to render a diagnosis decreased from 139.7 to 70.5 seconds. The authors concluded that this study demonstrates the value of applying a real-time AI solution to WSI to enhance the accuracy and efficiency of an intraoperative diagnosis of Hirschsprung disease.
Demir D, Ozyoruk KB, Durusoy Y, et al. The future of surgical diagnostics: AI-enhanced detection of ganglion cells for Hirschsprung disease. Lab Invest. 2025. doi.org/10.1016/j.labinv.2024.102189
Correspondence: Dr. Kutsev Bengisu Ozyoruk at [email protected], or Feras Alaqad at [email protected], or Dr. Mehmet Turan at [email protected]
Optimization of CPT coding in complex genitourinary surgical specimens
Pathology labs may find it difficult to apply Current Procedural Terminology codes to complex surgical specimens. The authors conducted a study at Emory University to assess and improve the accuracy of CPT coding for complex genitourinary specimens, specifically nephrectomy and cystectomy specimens. The study involved a baseline review of 71 specimens from April to June 2021, which revealed that 46 percent of the specimens were undercoded, primarily those with two or more billable organs. To address this issue, the institution implemented several changes, focusing on awareness of the problem, education, and documentation and communication involving pathology and billing teams. A subsequent review of 62 specimens from April to June 2023 showed significant improvement, with undercoding reduced to 21 percent. The study identified two main factors contributing to inaccurate CPT coding—incomplete documentation and lack of understanding of when tissues and organs should be bundled or unbundled for billing. One of the primary challenges involved nephrectomy specimens with commingled adrenal glands, for which proper documentation of the adrenal glands in the final diagnosis heading was essential for accurate coding. Additionally, the study highlighted the importance of recognizing when organs like the kidney and ureter should be billed separately, such as with nephroureterectomy specimens for urothelial carcinoma. Emory achieved these improvements in CPT coding through a collaborative effort between pathologists, laboratory assistants, pathology residents, and the billing team that emphasized the need for complete documentation and knowledge of CPT coding rules. The success of the project demonstrated that ongoing education and communication among all parties involved are crucial for optimizing coding for complex surgical specimens.
Behrman DB, Achram R, McClure C, et al. Optimization of Current Procedural Terminology coding in complex genitourinary surgical specimens. Arch Pathol Lab Med. 2025. doi.org/10.5858/arpa.2024-0118-OA
Correspondence: Dr. Lara Harik at [email protected]