Context.— Lymph node (LN) assessment plays a critical role in cancer staging and prognosis but remains a time-consuming and labor-intensive task in pathology. While artificial intelligence (AI) tools have shown promise in improving diagnostic accuracy, their real-world clinical utility in LN metastasis detection across multiple cancer types remains underexplored. Objective.— To evaluate the diagnostic performance and efficiency of an AI module in detecting LN metastases from gastric, colorectal, and breast cancers, and to assess its impact on pathologists’ workflow. Design.— A retrospective study was conducted by using 314 whole slide images from 95 patients who underwent resection for gastric, colorectal, or breast cancer. Three board-certified pathologists reviewed the slides with and without AI assistance. Diagnostic accuracy, review time, and number of mouse clicks required to detect metastases were recorded and compared. Results.— AI assistance increased sensitivity—which ranged from 91.8% to 93.9%—to 95.9% for all pathologists, while specificity remained high (97.0%–98.9%). Time to detect LN metastases decreased by up to 78% for some cancer types. The AI-guided click-based review required an average of 1.4 to 5.2 clicks depending on tissue type, with colorectal metastases detected most efficiently. Challenging subtypes, such as breast carcinoma with apocrine differentiation, required more extensive interaction. Micrometastases across all 3 cancer types were successfully identified by the AI. Conclusions.— The AI module improved pathologists’ sensitivity in detecting LN metastases and significantly reduced review time, particularly for positive nodes. These findings support the integration of AI tools to enhance diagnostic efficiency and accuracy in routine pathology practice.
Lami et al. (Mon,) studied this question.