In a Chinese study reported in The Lancet Oncology, Wu et al found that an artificial intelligence (AI)-based lymph node metastases diagnostic model using whole-slide images performed well in identifying lymph node metastases in patients with bladder cancer, as well as other cancers.
Study Details
The retrospective multicenter study included 998 patients who had undergone radical cystectomy and pelvic lymph node dissection between January 2013 and December 2021 and had whole-slide images of lymph node sections available; 7,991 whole-slide images were included in the study.
The population comprised a training set and five validation sets. A subset of challenging cases from the validation sets was used to compare performance between the lymph node metastases diagnostic model and pathologists. The lymph node metastases diagnostic model was also tested in a breast cancer image dataset and a prostate cancer image dataset.
Key Findings
The receiver operating characteristic area under the curve (AUC) values for diagnosis with the lymph node metastases diagnostic model ranged from 0.978 (95% confidence interval [CI] = 0.960–0.996) to 0.998 (95% CI = 0.996–1.000) in the five validation sets. The lymph node metastases diagnostic model detected tumor micrometastases in 13 patients in the validation sets categorized as negative for metastasis by pathologists. When retaining 100% sensitivity, the lymph node metastases diagnostic model would have enabled pathologists to exclude 80% to 92% of negative slides across the validation sets.
Comparisons between the lymph node metastases diagnostic model and pathologists showed that the diagnostic sensitivity of the model (0.983, 95% CI = 0.941–0.998) exceeded that of junior (0.906, 95% CI = 0.871–0.934) and senior pathologists (0.947, 95% CI = 0.919–0.968). AI assistance improved sensitivity for junior pathologists to 0.953 and for senior pathologists to 0.986.
The lymph node metastases diagnostic model exhibited AUC values of 0.943 (95% CI = 0.9180.969) in the breast cancer image dataset and 0.922 (95% CI = 0.884–0.960) in the prostate cancer image dataset.
The investigators concluded, “We developed an AI-based diagnostic model that did well in detecting lymph node metastases, particularly micrometastases. The lymph node metastases diagnostic model showed substantial potential for clinical applications in improving the accuracy and efficiency of pathologists’ work.”
Tianxin Lin, PhD, of the Department of Urology, Sun Yat-sen Memorial Hospital, Sun Yat-sen University, Guangzhou, is the corresponding author for The Lancet Oncology article.
Disclosure: The study was funded by the National Natural Science Foundation of China, Science and Technology Planning Project of Guangdong Province, and others. For full disclosures of the study authors, visit thelancet.com.
