Predicting response to neoadjuvant chemotherapy for colorectal liver metastasis using deep learning on prechemotherapy cross-sectional imaging.

BACKGROUND AND OBJECTIVES: Deep learning models (DLMs) are applied across domains of health sciences to generate meaningful predictions. DLMs make use of neural networks to generate predictions from discrete data inputs. This study employs DLM on prechemotherapy cross-sectional imaging to predict patients' response to neoadjuvant chemotherapy. METHODS: Adult patients with colorectal liver metastasis who underwent surgery after neoadjuvant chemotherapy were included. A DLM was trained on computed tomography images using attention-based multiple-instance learning. A logistic regression model incorporating clinical parameters of the Fong clinical risk score was used for comparison. Both model performances were benchmarked against the Response Evaluation Criteria in Solid Tumors criteria. A receiver operating curve was created and resulting area under the curve (AUC) was determined. RESULTS: Ninety-five patients were included, with 33,619 images available for study inclusion. Ninety-five percent of patients underwent 5-fluorouracil-based chemotherapy with oxaliplatin and/or irinotecan. Sixty percent of the patients were categorized as chemotherapy responders (30% reduction in tumor diameter). The DLM had an AUC of 0.77. The AUC for the clinical model was 0.41. CONCLUSIONS: Image-based DLM for prediction of response to neoadjuvant chemotherapy in patients with colorectal cancer liver metastases was superior to a clinical-based model. These results demonstrate potential to identify nonresponders to chemotherapy and guide select patients toward earlier curative resection.

10 years, 100 robotic major hepatectomies: a single-center experience.

INTRODUCTION: Adoption of robotic liver resections has been gradually increasing throughout the HPB surgical community over the past decade. Currently there is limited literature which demonstrates a significant benefit of robotic surgery for major hepatectomies over open or laparoscopic. As one of the first centers to develop a robotic HPB program, we have experienced improved outcomes over time with increasing utilization of robotics. Herein, we present our 10-year experience and outcomes for major robotic liver resections. METHODS: From 2012 to 2022, 361 robotic liver procedures were performed, including 100 major hepatectomies. A retrospective data review of the electronic medical record was performed evaluating outcomes after robotic major hepatectomy. Outcomes for the first 50 cases (Group A) and second 50 cases (Group B) were compared to identify any improvements in practice. Demographic and clinical outcome variables were analyzed. Data were assessed for normality, and Wilcoxon rank-sum, χ2 tests, and a logistic regression model were performed appropriate for the data. Stata v.17 was utilized, and significance was set as p < .05. RESULTS: There was no difference in median operative time (258 vs 256 min), EBL (500 vs 500 mL), median LOS (5 vs 3.5 days), 90-day readmission (14% vs 24%), major complications (14% vs 20%), and 90-day mortality (6% vs 4%) between early and late cases, respectively. ICU admissions and conversion rates were significantly lower in group B (14.0% vs 48.0%), while expert level difficulty indices were higher (82% vs 58%). CONCLUSION: Development of a robotic liver program with good outcomes is feasible over time. Our data suggest that our institutional learning curve for robotic major hepatectomy plateaued at approximately 50 cases.

Expanding the utility of robotics for pancreaticoduodenectomy: a 10-year review and comparison to international benchmarks in pancreatic surgery.

BACKGROUND: Robotic pancreaticoduodenectomy (RPD) is an emerging alternative to open pancreaticoduodenectomy (OPD). Although RPD offers various theoretical advantages, it is used in less than 10% of all pancreaticoduodenectomies. The aim of this study was to report our 10-year experience and compare RPD outcomes with international benchmarks for OPD. METHODS: A retrospective review of a prospectively maintained institutional database was performed of consecutive patients who underwent RPD between January 2011 and December 2021. Patients were categorized into low-risk and high-risk groups according to the selection criteria set by the benchmark study. Their outcomes were compared to the international benchmark cut off values. Outcomes were then evaluated over time to identify improvements in practice and establish a learning curve. RESULTS: Of 201 RPDs, 36 were low-risk and 165 high-risk patients. Compared to the OPD benchmarks, outcomes of low-risk patients were within the cutoff values. High-risk patients were outside the cutoff for blood transfusions (26% vs. ≤ 23%), overall complications (78% vs. ≤ 73%), grade I-II complications (68% vs. ≤ 62%), and readmissions (22% vs ≤ 21%). Oncologic outcomes for high-risk patients were within benchmark cutoffs. Cases at the end of the learning curve included more pancreatic cancer (42% from 17%) and fewer low-risk patients (10% from 24%) than those at the beginning. After 41 RPD there was a decline in conversion rates and operative time. Between 95 and 143 cases operative time, transfusion rates, and LOS declined significantly. Complications did not differ over time. CONCLUSION: RPD yields results comparable to the established benchmarks in OPD in both low- and high-risk patients. Along the learning curve, RPD evolved with the inclusion of more high-risk cases while outcomes remained within benchmarks. Addition of a robotic HPB surgery fellowship did not compromise outcomes. These results suggest that RPD may be an option for high-risk patients at specialized centers.