Artificial intelligence-based model for lymph node metastases detection in bladder cancer.

Abstract

Artificial intelligence can reach the level of trained pathologists in terms of diagnostic accuracy, thereby substantially reducing the clinical workload of front-line pathologists in the future. 1 Campanella G Hanna MG Geneslaw L et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images. Nat Med. 2019; 25: 1301-1309 Crossref PubMed Scopus (847) Google Scholar However, artificial intelligence-based models require large-scale validation in real-world clinical scenarios. 1 Campanella G Hanna MG Geneslaw L et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images. Nat Med. 2019; 25: 1301-1309 Crossref PubMed Scopus (847) Google Scholar , 2 Ström P Kartasalo K Olsson H et al. Artificial intelligence for diagnosis and grading of prostate cancer in biopsies: a population-based, diagnostic study. Lancet Oncol. 2020; 21: 222-232 Summary Full Text Full Text PDF PubMed Scopus (238) Google Scholar In their Article, 3 Wu S Hong G Xu A et al. Artificial intelligence-based model for lymph node metastases detection on whole slide images in bladder cancer: a retrospective, multicentre, diagnostic study. Lancet Oncol. 2023; 24: 360-370 Summary Full Text Full Text PDF PubMed Scopus (2) Google Scholar Shaoxu Wu and colleagues conducted a large-scale, multicentre, retrospective study on the use of an artificial intelligence-based model for the diagnosis of lymph node metastases in bladder cancer. Their results showed that the diagnostic sensitivity of the model (0·983 [95% CI 0·941–0·998]) substantially exceeded that of both junior pathologists (0·906 [0·871–0·934]) and senior pathologists (0·947 [0·919–0·968]). Regarding the multicancer test, the model also demonstrated excellent diagnostic performance (area under the curve of 0·943 [95% CI 0·918–0·969] in breast cancer images and 0·922 [0·884–0·960] in prostate cancer images). The lymph node metastases diagnostic model can exclude at least 80% of examination tasks while maintaining 100% sensitivity in clinical applications. These results are encouraging, and we sincerely congratulate Wu and colleagues on their high-quality research. We also propose some questions to help us better understand this study and address confusion in our own research on artificial intelligence. Artificial intelligence-based model for lymph node metastases detection on whole slide images in bladder cancer: a retrospective, multicentre, diagnostic studyWe developed an AI-based diagnostic model that did well in detecting lymph node metastases, particularly micrometastases. The LNMDM showed substantial potential for clinical applications in improving the accuracy and efficiency of pathologists' work. Full-Text PDF Artificial intelligence-based model for lymph node metastases detection in bladder cancer – Authors' replyWe appreciate the Correspondences from Zhipeng Mai and colleagues and Kang Zou and colleagues on our retrospective, multicentre study developing an artificial intelligence-based model for the detection of lymph node metastases on whole slide images in bladder cancer.1 Full-Text PDF Artifical intelligence-based model for lymph node metastases detection in bladder cancerWe read with interest the Article by Shaoxu Wu and colleagues,1 which reports on the development of a diagnostic model based on pathomics for the detection of lymph node metastases in bladder cancer. Although the authors stated that their artificial intelligence-based model could achieve 100% sensitivity with an acceptable false positive rate, lowering the diagnostic threshold to achieve high sensitivity is bound to sacrifice the corresponding positive predictive value (PPV). Among the five validation sets, the PPVs of three sets were less than satisfactory (ie, <0·700), especially in medical centres with a low proportion of patients with lymph node metastases-positive images, such as the Third Affiliated Hospital of Sun Yat-sen University, which reported a PPV of only 0·283. Full-Text PDF