25 January 2024 in NEJM AI
1 February 2017 in Journal of Clinical Oncology
697 Background: A major challenge in identifying candidates for nonoperative management of locally advanced rectal cancer is predicting pathological complete response (pCR) following chemoradiation therapy (CRT). We evaluated the ability of pre- and post-CRT PET imaging to predict pCR and long-term prognosis. Methods: We retrospectively identified patients at our institution from 2002–2015 with locally advanced rectal cancer who underwent CRT, pre- and post-CRT PET imaging, and surgical resection. Logistic regression and Kaplan-Meier estimates were used to analyze the association of PET variables with tumor pCR and survival outcomes. Receiver operator characteristic curves were generated to define optimal cutoff points for predictive PET variables. Results: 140 patients matched the inclusion criteria. The pCR rate was 28%, and median follow-up time was 48 months. On multivariable analysis, pCR was associated with lower median post-CRT SUVmax(3.2 vs 5.2, p=0.009) and higher median SUV percent decrease (72 vs 58%, p=0.009). ROC curves were generated for SUV percent decrease (AUC=0.70) and post-CRT SUV (AUC=0.69) to estimate cutoff points for maximum specificity and sensitivity to predict pCR. Post-CRT SUV 66% were equally predictive of pCR with a sensitivity of 65%, specificity of 72%, PPV of 44%, and NPV of 86%. Median 5-year OS and RFS were significantly improved for patients with post-CRT SUV 66% (OS, 82 vs 66%, p=0.05; RFS, 75 vs 54%, p=0.01). Patients with stage III disease and a post-CRT SUV <4.3 were in effect downstaged, with a median 5-year OS equivalent to that of patients with stage II disease (Table 1; 86 vs 86%). Conclusions: PET/CT may be a useful modality in predicting pCR and overall survival in patients undergoing CRT for rectal cancer. Inclusion of pre-CRT PET does not appear to add prognostic value for pCR compared with post-CRT PET alone. Patients with a post-CRT SUV of >4.3 should not be considered for nonoperative management, as an estimated 86% of these patients will not have a pCR.
27 November 2025 in medRxiv
Background: Traditional surgical training relies on an apprenticeship model, which is subjective and threatened by human bias. Performance metric scales attempt to offer more objective feedback by providing a structured grading rubric, but these scores are still ultimately subjective. Leveraging computer vision and artificial intelligence to assess surgical performance has the potential to shift subjective measurements into automated and objective feedback for trainees. Materials and Methods: This retrospective, multi-institutional study analyzed 319 laparoscopic cholecystectomy videos from IRB-approved deidentified datasets and segmented the videos into 2862 clips. Using an internally validated video-based assessment rubric, we annotated video clips across five metacompetency domains: tissue handling, psychomotor skills, efficiency, dissection quality, and exposure quality. Short video segments ( < 90s) were rated on a 5-point scale by expert raters. We trained a deep learning model (DINOv2) to classify composite high (4–5) vs low (1–3) metacompetency scores, representing yes/no binary feedback, a realistic comparison to the operating room. Model performance was evaluated via area under the receiver operating characteristic curve. Results: Among 2862 LC video clips, model performance was highest for dissection quality during the exposing gallbladder step (AUROC 91.5%, 95% confidence interval [CI], 84.5-96.5). Moderate performance was observed for efficiency (AUROC 72.6%, 95% CI 59.9-83.2) and exposure quality (AUROC 68.7%, 95% CI 55.2-81.8). Dissection and exposure quality scores during hepatocystic triangle dissection yielded AUROCs of 63.8% (95% CI 56.9-71.3) and 66.0% (95% CI 53.1-76.7), respectively. Conclusion: We demonstrate the feasibility of a purely vision-based deep learning model to grade surgical skill based on metacompetencies with excellent performance during simple steps. This technique represents an advance over prior whole-video approaches that rely heavily on tool- tracking and kinematic data, and may lead to greater model performance by using binary feedback on increasingly specific step segmentation.