Postma, Alida A.

The translation of AI-generated brain metastases (BM) segmentation into clinical practice relies heavily on diverse, high-quality annotated medical imaging datasets. The BraTS-METS 2023 challenge has gained momentum for testing and benchmarking algorithms using rigorously annotated internationally compiled real-world datasets. This study presents the results of the segmentation challenge and characterizes the challenging cases that impacted the performance of the winning algorithms. Untreated brain metastases on standard anatomic MRI sequences (T1, T2, FLAIR, T1PG) from eight contributed international datasets were annotated in stepwise method: published UNET algorithms, student, neuroradiologist, final approver neuroradiologist. Segmentations were ranked based on lesion-wise Dice and Hausdorff distance (HD95) scores. False positives (FP) and false negatives (FN) were rigorously penalized, receiving a score of 0 for Dice and a fixed penalty of 374 for HD95. The mean scores for the teams were calculated. Eight datasets comprising 1303 studies were annotated, with 402 studies (3076 lesions) released on Synapse as publicly available datasets to challenge competitors. Additionally, 31 studies (139 lesions) were held out for validation, and 59 studies (218 lesions) were used for testing. Segmentation accuracy was measured as rank across subjects, with the winning team achieving a LesionWise mean score of 7.9. The Dice score for the winning team was 0.65 ± 0.25. Common errors among the leading teams included false negatives for small lesions and misregistration of masks in space. The Dice scores and lesion detection rates of all algorithms diminished with decreasing tumor size, particularly for tumors smaller than 100 mm3. In conclusion, algorithms for BM segmentation require further refinement to balance high sensitivity in lesion detection with the minimization of false positives and negatives. The BraTS-METS 2023 challenge successfully curated well- annotated, diverse datasets and identified common errors, facilitating the translation of BM segmentation across varied environments and providing the tools for future development of personalized volumetric reports to patients undergoing BM treatment.

Importance A subset of Union for International Cancer Control ( UICC )/ American Joint Committee on Cancer ( AJCC ) eighth edition TNM stage I and II human papillomavirus–positive oropharyngeal carcinoma has undesirable outcomes, which might have contributed to a lack of success in phase III deintensification trials. Refining clinical stage groups, especially in the overabundant cN1/stage I group, has become important for treatment selection. Objective To assess the prognostic importance of pretreatment lymph node (LN) characteristics to optimize case distribution and outcome homogeneity within the N classification system. Design, Setting, and Participants This is an international multi-institutional retrospective prognostic cohort study. Analysis of human papillomavirus–positive oropharyngeal carcinoma treated curatively from 4 institutions (International Collaboration of Oropharyngeal Cancer Network for N-Classification [ICON-N] dataset) provided a refined clinical staging proposal; an independent dataset (Centre Hospitalier de l’Université de Montréal [CHUM] dataset) validated the proposal. Neuroradiologists reviewed pretreatment computed tomography and/or magnetic resonance imaging for nodal features, including presence or absence of abnormal LN(s), retropharyngeal LN, laterality, number of abnormal LN, and imaging-detected extranodal extension (iENE). Data were collected from February to May 2023, and data were analyzed from June to July 2023. Exposures Definitive chemoradiotherapy/radiotherapy or definitive surgery with or without postoperative chemoradiotherapy/radiotherapy. Main Outcomes and Measures The primary end point was overall survival. A Cox proportional hazards multivariable model was used to estimate adjusted hazard ratios (AHRs) and to derive an optimal clinical TNM stage classification (AHR-stage schema) incorporating the strongest prognostic nodal features within the UICC/AJCC eighth edition TNM framework after confirming the prognostication of iENE status. The performance (according to overall normalized scores and ranking) of the AHR-stage schema against the current UICC/AJCC eighth edition TNM staging system was evaluated for hazard consistency, hazard discrimination, prognostic importance, and sample size balance. Validation was performed in the CHUM dataset. Results The ICON-N dataset comprised 2053 patients, including 1898 (92.5%) with cN-positive disease and 155 (7.5%) with cN0 disease; a total of 298 (14.5%) were female, and the mean (SD) age was 60.6 (9.3) years. iENE-positive disease was identified in 710 of 1898 patients with cN-positive disease (37.4%). The median (range) follow-up was 5.1 (0.1-14.7) years. iENE was the strongest prognostic nodal feature in multivariable analysis; the AHR for iENE-positive vs iENE-positive disease was 2.43 (95% CI, 1.96-3.03) in the ICON-N dataset and 2.04 (95% CI, 1.28-3.23) in the CHUM dataset (n = 451). Reclassifying iENE-positive cases 1 stratum higher for N categorization without altering iENE-negative cases yielded an AHR-stage schema that outperformed the current TNM staging system in disease-free and overall survival with a lower (ie, better) overall normalized score (2 vs 3). Conclusions and Relevance In this study, reclassifying each N category 1 stratum higher for iENE-positive disease resulted in better disease-free and overall survival. The proposed new classification outperformed the currently TNM staging system in risk stratification and may facilitate future clinical trial design, outcomes research, and patient care.