Comparative Diagnostic Performance of Endoscopic Classifications for Predicting Histopathology in Large Laterally Spreading Colorectal Tumors

Introduction. Colorectal cancer is a major cause of cancer-related mortality worldwide, and early, accurate diagnosis is essential for effective management. Optical diagnosis using image-enhanced endoscopy and standardized classifications has improved real-time assessment of lesion morphology, yet the optimal approach for large laterally spreading tumors (LSTs), especially granular mixed type, remains under debate. This study aimed to compare the diagnostic performance of the JNET, Kudo, Modified Sano, Hiroshima classifications, and forceps biopsy in predicting histopathology for large LSTs, with a focus on differentiating between granular (LST-G) and non-granular (LST-NG) subtypes. Methods. Ninety-five patients with LSTs > 20 mm were enrolled, with the largest lesion per patient selected for analysis. Patients were stratified into LST-G and LST-NG groups using the Paris classification. Lesions were evaluated using optical diagnosis based on four endoscopic classifications: Kudo, JNET, Modified Sano, and Hiroshima. Targeted forceps biopsies were obtained from areas of the most pronounced changes, and subsequent endoscopic resection specimens served as the histopathological reference standard. The diagnostic performance metrics were calculated, including sensitivity, specificity, predictive values, and diagnostic accuracy. Results. In the LST-G group, the Modified Sano classification achieved the highest sensitivity (92.31%) but had low specificity (52.38%). In contrast, the Hiroshima classification exhibited low sensitivity (65.22%) but achieved superior specificity (100%), along with the highest positive predictive value (100%), negative predictive value (82.22%), and overall diagnostic accuracy (86.67%). Forceps biopsy demonstrated a balanced diagnostic performance, slightly surpassing JNET in sensitivity (79.49% vs. 71.79%), specificity (95.24% vs. 90.48%), negative predictive value (96.88% vs. 93.33%), positive predictive value (71.43% vs. 63.33%), and diagnostic accuracy (85% vs. 78.3%). In contrast, for LST-NG lesions, the JNET classification outperformed other modalities across most diagnostic metrics, with the highest sensitivity (71.43%), specificity (100%), and positive predictive value (100%); however, it had a slightly lower negative predictive value and diagnostic accuracy compared to the Hiroshima classification (84% vs. 93.75% and 88.57% vs. 94.29%, respectively). Conclusions. Among the evaluated endoscopic classifications, JNET proved to be the most effective for large LST-NG, while for LST-G, the Modified Sano classification demonstrated the highest sensitivity, and the Hiroshima classification excelled in other diagnostic metrics. Although further improvements in optical diagnosis are warranted, targeted forceps biopsy provides no additional diagnostic benefits before resection.