Masashi Takeuchi

Objective: To develop an artificial intelligence (AI) system for anatomical recognition that can automatically identify key anatomical structures during bladder neck dissection in robot‐assisted radical prostatectomy (RARP), a critical yet technically demanding step. Materials and Methods: We constructed an AI model and evaluated its accuracy using 210 RARP videos from two institutions. A total of 1353 frames of bladder neck dissection were extracted from 25 videos, and the boundaries of the prostate, bladder and retrotrigonal layer were annotated. The DeepLabV3+ model was employed in developing the anatomical recognition AI model. An active learning approach was applied by iteratively retraining the model using frames with poor initial performance. The model was further evaluated using 202 independent test frames derived from seven surgeons. Segmentation performance was evaluated using the intersection over union (IoU) and Dice similarity coefficient. Results: The IoU of the initial AI model was 0.41 for the prostate and 0.44 for the bladder. The retrained model demonstrated improved segmentation performance, with final IoU values of 0.75 and 0.68, respectively. In an independent test dataset derived from seven surgeons, segmentation performance for the prostate and bladder was maintained. Conclusion: We developed an AI system for anatomical recognition and improved its accuracy by using retraining processes. The retrained AI model demonstrated high accuracy, and its use is anticipated to support surgeons and aid in the education of novice surgeons.

Background Artificial intelligence (AI) has rapidly advanced in surgical applications. However, existing single-modality AI models relying solely on image input lack the ability to integrate anatomical understanding with clinical reasoning, which is essential for safe and actual surgical decision-making. We constructed an AI model with multimodal training combining visual and linguistic data and named Surgical Vision-Language Model (Surgical-VLM) for real-time surgical support. Methods We analyzed surgical videos from 50 cases of robotic distal gastrectomy and extracted 50 still images per case, generating 10,000 vision–question–answer (VQA) pairs. The model was fine-tuned using the Large Language and Vision Assistant (LLaVA) framework. Model performance was assessed using the newly developed Surgical-VLM Bench, which evaluates appropriateness of expression, anatomical accuracy, and clinical usefulness on a 5-point scale, and the Bidirectional Encoder Representations from Transformers (BERT) score. The results were compared with those of ChatGPT-5. Results Surgical-VLM achieved a mean total benchmark score of 11.33 ± 0.697 versus 10.72 ± 0.536 for ChatGPT-5. Surgical-VLM showed numerically higher mean scores in anatomical accuracy and clinical usefulness. The BERTScore was 0.768 ± 0.008 for Surgical-VLM and 0.804 ± 0.013 for ChatGPT-5. Conclusions This proof-of-concept study demonstrates the feasibility of domain adaptation for a Surgical-VLM and proposes a clinically grounded benchmark for structured evaluation. In this pilot setting, the prototype generated context-aware responses and showed domain-dependent differences compared with a general-purpose multimodal model. Further validation with larger independent test sets, expanded VQA items, and external evaluators is required before any clinical use.

Artificial intelligence (AI) has shown potential in various fields; however, its practical application in surgery remains limited. We developed an AI syste

Background Precise recognition of anatomical structures such as the pancreas, transverse mesocolon, and right gastroepiploic vein (RGEV) is critical for the safety and effectiveness of infrapyloric lymph node dissection during robotic distal gastrectomy (RDG). Misrecognizing these structures can increase the risk of complications. However, even expert surgeons may encounter difficulties identifying these structures due to variations in patient anatomy and surgical technique. This study aims to validate the performance of an artificial intelligence (AI) model for automated anatomical recognition during RDG across multiple institutions and evaluate its clinical utility in supporting surgeons. Methods Surgical videos from 90 patients who underwent RDG at multiple high-volume centers in Japan were analyzed. The AI model’s accuracy was evaluated using Intersection over Union (IoU) for the pancreas, transverse mesocolon, and RGEV across four high-volume institutions. The clinical utility of the AI model was assessed by comparing the accuracy of surgeons in identifying anatomical landmarks during infrapyloric lymph node dissection between procedures performed with and without AI assistance. Results During this surgical procedure, the IoU of our AI model for the pancreas, transverse mesocolon, and RGEV were 0.624, 0.575, and 0.618, respectively. Furthermore, despite differences in accuracy across the four facilities, the AI model achieved consistent IoU values ranging from 0.5 to 0.7 across all institutions, demonstrating minimal variability. AI-assisted surgeons showed significantly higher accuracy in recognizing anatomical landmarks, particularly the transverse mesocolon’s edge, compared with those without AI assistance (91.7% vs. 80.6%, p = 0.043). These findings highlight the model’s ability to enhance the safety and precision of infrapyloric lymph node dissection. Conclusion The AI model demonstrated strong external validity and improved surgeons’ recognition of anatomical structures during RDG. Its integration into clinical practice can potentially reduce the rate of surgical complications and improve patient outcomes. Graphical Abstract

Nakano and colleagues developed an artificial intelligence‐enhanced navigation system for robotic hepatectomy, enabling early identification of the inferior vena cava and major hepatic vein roots. Using semantic segmentation on 1030 annotated images, the model showed reliable performance and may help reduce complications, enhance safety, and support minimally invasive liver surgery.

As a first step to prevent recurrent laryngeal nerve (RLN) palsy, we have developed an artificial intelligence (AI)-based anatomical recognition system for

Recurrent laryngeal nerve (RLN) palsy often occurs due to excessive traction (ET) on the nerve during esophagectomy. Use of a nerve integrity monitor (NIM) sometimes can prevent RLN palsy, but it does not detect injury before it occurs. The aim of this proof-of-concept study was to use an artificial intelligence (AI) system for pre-injury detection of ET on the left RLN during robot-assisted minimally invasive esophagectomy (RAMIE). We extended a previously developed AI model for anatomical recognition by labeling video frames from 130 RAMIE patients as ET or non-ET according to visual indicators of nerve tension. The system quantifies the probability of ET in real-time as an excessive traction risk (ETR), which is displayed on the surgical video. The AI correctly identified 84.4% of scenes involving unintended nerve traction in 10 surgical patients, and the ETR showed a correspondence with the actual degree of traction applied. Notably, in a representative patient, the AI detected ET earlier than the decrease in the NIM amplitude. By providing real-time proactive feedback, the system enables earlier recognition of hazardous traction, which may allow adjustment before irreversible nerve damage occurs.

Aim: Previous studies have shown that the volume and short‐term outcomes of advanced hepatectomy in Japan remained stable during the coronavirus disease 2019 (COVID‐19) pandemic. However, whether these trends have changed in the postpandemic period remains unclear. This study aimed to evaluate surgical volume and short‐term outcomes following advanced hepatectomy in Japan during the postpandemic era. Methods: Data from the Japanese National Clinical Database (NCD) were analyzed for patients who underwent advanced hepatectomy between 2018 and 2023. Changes in the number of the procedures, major complications (Clavien–Dindo grade ≥ III), 30‐day and inhospital mortality rates, and failure‐to‐rescue rates were assessed. The standardized morbidity and mortality ratios—calculated as the observed‐to‐expected incidence rates using an NCD‐established risk model for 30‐day mortality, inhospital mortality, and major complications—were also examined. Results: A total of 39 348 cases were included. The number of advanced hepatectomies showed a gradual decline, independent of the COVID‐19 pandemic. However, the proportion of patients aged over 80 years significantly increased throughout the study period. Monthly standardized mortality and morbidity ratios largely remained stable across the study period, including during the pandemic and postpandemic eras. Conclusions: Analysis of data from a nationwide Japanese database indicates that advanced hepatectomy continues to be performed safely in the post‐COVID‐19 era, despite a decreasing procedural volume.