Multiservice machine learning models were developed to predict postsurgical length of stay (LOS) and discharge disposition at the time of case posting. An analysis of 63,574 patients showed the LOS model achieved an area under the curve (AUC) of 0.81. Incorporating relative value units and historical LOS improved prediction accuracy for both short and prolonged […]
Category: Digital Surgery and Telemedicine
Machine learning effectively predicts discharge against medical advice
A study evaluated machine learning algorithms to predict discharge against medical advice for 48,394 injured inpatients over five years. Among these individuals, 8.8% opted for discharge against medical advice. The light gradient boosting machine combined with edited nearest neighbors outperformed traditional logistic regression, achieving areas under the curve of 0.820 and 0.837 in internal and […]
Machine learning predicts surgical outcomes in cholangiocarcinoma patients
A machine learning model developed from 376 intrahepatic cholangiocarcinoma patients effectively predicts textbook outcomes. Key preoperative factors include Child-Pugh classification, ECOG score, hepatitis B status, and tumor size. The model achieved high accuracy during internal (AUC = 0.8825) and external validation (AUC = 0.8346). Survival analysis indicated that patients achieving textbook outcomes had disease-free survival […]
Deep learning model enhances bile duct identification during surgery
A newly developed deep learning system effectively identifies extrahepatic bile ducts in real-time during laparoscopic cholecystectomy. Trained on 3,993 images, the YoloV7 model achieved a mean average precision of 0.846 overall, with specific accuracies of 94.39% for the common bile duct and 84.97% for the cystic duct during video clip validations. By optimizing these crucial […]
Automated models enhance competency assessment in laparoscopic surgery
The study demonstrates the development of automated surgical action recognition models, achieving significant accuracy in competency prediction during laparoscopic cholecystectomy. Analysis of the cholec80 dataset revealed that high-competency groups exhibited shorter dissection durations and higher scores on established evaluation metrics. A random forest model achieved 93% accuracy in predicting surgical competency, while a video-masked autoencoder […]
Prognostic model predicts recurrence and treatment response in HCC
A novel prognostic model utilizing pathological signatures effectively predicts postoperative recurrence risk and sorafenib response in hepatocellular carcinoma (HCC) patients. Analyzed across 287 non-treated patients, the model achieved AUROC values of 0.818 and 0.811 for predicting one and two-year recurrence respectively. Validation with an external cohort confirmed its predictive power. Additionally, it successfully stratified sorafenib-treated […]
Large language models can improve surgical patient education
A multicenter analysis revealed that ChatGPT-4o effectively enhances patient education before and after surgery. Evaluating audio responses to frequently asked questions, the model achieved an average accuracy score of 4.12/5 and a relevance score of 4.46/5. Postoperative responses were notably more accurate and less harmful than preoperative ones. Results suggest that integrating ChatGPT-4o in clinical […]
Machine learning effectively predicts postpancreatectomy acute pancreatitis
A novel machine learning model successfully predicts postpancreatectomy acute pancreatitis (PPAP) in patients following pancreaticoduodenectomy. In a cohort of 381 patients, 88 (23.09%) developed PPAP, which was notably associated with a higher occurrence of postoperative pancreatic fistulas (55.68%). Various algorithms, including logistic regression and gradient boosting, were tested, with recursive feature elimination optimizing variable selection. […]
Improved visualization of surgical needle tips enhances accuracy
A novel model was developed to visualize hidden surgical needle tips within organs, addressing challenges in laparoscopic procedures that could lead to complications like anastomotic leakage. Testing revealed real-time image inference at 33.4 frames per second and a mean needle misalignment of just 1.03 mm, well below the 1.8 mm threshold. These findings suggest that […]
AI has transformative potential for surgical education.
Artificial intelligence could revolutionize surgical education by providing personalized feedback, improved competency evaluations, and enhanced candidate selection. AI-driven simulations foster adaptive learning among trainees, while intraoperative tools may assist surgeons in complex procedures. However, challenges such as data quality, ethical concerns, and the risk of overskilling need to be addressed. Developing regulatory frameworks emphasizing transparency […]