Deep Learning HRNet-FCN for Blood Vessel Identification in Laparoscopic Pancreatic Surgery

Deep Learning HRNet-FCN for Blood Vessel Identification in Laparoscopic Pancreatic Surgery | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Help Center Sign In Submit a Preprint Cite Share Download PDF Article Deep Learning HRNet-FCN for Blood Vessel Identification in Laparoscopic Pancreatic Surgery Jile Shi, Ruohan Cui, Zhihong Wang, Qi Yan, Lu Ping, Hu Zhou, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5472618/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 01 May, 2025 Read the published version in npj Digital Medicine → Version 1 posted 11 You are reading this latest preprint version Abstract Laparoscopic pancreatic surgery remains highly challenging due to the complexity of the pancreas and surrounding vascular structures, with risk of injuring critical blood vessels such as the Superior Mesenteric Vein (SMV)-Portal Vein (PV) axis and splenic vein. Here, we evaluated the High Resolution Network (HRNet)-Full Convolutional Network (FCN) model for its ability to accurately identify vascular contours and improve surgical safety. Using 12,694 images from 126 laparoscopic distal pancreatectomy (LDP) videos and 35,986 images from 138 Whipple procedure videos, the model demonstrated robust performance, achieving a mean Dice coefficient of 0.754, a recall of 85.00%, and a precision of 91.10%. By combining datasets from LDP and Whipple procedures, the model showed strong generalization across different surgical contexts and achieved real-time processing speeds of 11 frames per second. These findings highlight the potential of HRNet-FCN to recognize anatomical landmarks, enhance surgical precision, reduce complications, and improve outcomes in laparoscopic pancreatic procedures. Biological sciences/Computational biology and bioinformatics/Image processing Biological sciences/Computational biology and bioinformatics/Machine learning Health sciences/Diseases/Gastrointestinal diseases/Pancreatic disease Health sciences/Diseases/Gastrointestinal diseases/Pancreatic disease/Pancreatic cancer Health sciences/Health care/Therapeutics/Surgery Health sciences/Health care/Therapeutics/Surgery/Surgical oncology Figures Figure 1 Figure 2 Figure 3 Introduction As an increasingly popular treatment for pancreatic diseases, laparoscopic pancreatic surgery offers a minimally invasive approach, reducing recovery time, postoperative pain and hospital stays 1 . However, laparoscopic pancreatic surgery remains highly challenging due to the complexity of the pancreas and surrounding structures. One of the main challenges for pancreatic surgery is the manipulation of critical anatomical vessels, especially the Superior Mesenteric Vein-Portal Vein (SMV-PV) axis and the splenic vein 2 . These vessels pose a significant challenge due to their susceptibility to intraoperative bleeding, exacerbated by the veins’ delicate nature 3 . SMV-PV axis is a crucial factor in determining the resectability of pancreatic tumors, especially when venous invasion occurs, requiring preservation or reconstruction for successful surgery 4 – 6 . Proper management of this axis during venous resection is essential to minimize complications and maintain splenic vein function, which significantly impacts surgical outcomes 7 . Therefore, accurate identification and careful handling of the SMV-PV axis and splenic vein is one of the most difficult complications in pancreatic surgery 4 . However, as the surgeon operating with the laparoscopy cannot use the "sense of touch" to identify blood 8 , it is important to enhance the visual identification of SMV-PV axis and splenic vein during laparoscopic pancreatic surgery. In the field of medical imaging recognition, Deep Learning (DL) technology has catalyzed significant breakthroughs in previous medical imaging studies, like ultrasound 9 , PET-CT 10 , CT 11 , MR 12 and Retinal Fundus Photographs 13 . Specifically, in the realm of intelligent surgery, DL has proven its merit in accurately identifying and segmenting critical arteries, such as renal 14 and mesenteric arteries 15 , achieving impressive precision of 0.937. Over the past three years, several leading publications have reported the application of DL technology in identifying anatomical landmarks and safety assessments in surgical procedures 16 – 20 . However, these studies predominantly focused on cholecystectomy and endoscopic pituitary surgery, leaving other surgical scenarios, such as pancreatic surgery, relatively unexplored. Here, we present two examples of SMV-PV axis and the splenic vein recognition in laparoscopic pancreatic surgery, including Laparoscopic Distal Pancreatectomy (LDP) and Pancreaticoduodenectomy (Whipple procedure). LDP and Whipple are both widely recognized as a standard procedure for both benign and malignant pancreatic diseases. LDP is particularly for lesions located in the pancreatic tail or body 21 – 24 , while the Whipple procedure is primarily performed for lesions located in the pancreatic head, as well as the duodenum, common bile duct, or surrounding areas 25 . However, both LDP and Whipple are complex surgical procedure that requires high level qualifications and trainings for surgeons: surgeons typically need to complete hundreds of laparoscopic procedures in other areas under to develop the expertise necessary for safe performance of LDP 26 and Whipple 27 , 28 . Therefore, it is important to enhance the venous anatomical landmarks for LDP Whipple procedure, for assisting surgery and educational purpose. In both processes, we constructed an annotated database of SMV-PV images from experienced surgeons. We then employed the High Resolution Network (HRNet) 29 to train the model, which identified and delineated the SMV-PV axis and splenic vein in LDP, and explored the real time segmentation of anatomical landmarks. Our model allows for instant identification of the SMV-PV axis during pancreatic surgery, enhancing surgical precision. Overall, this work can enhance the safety, reduces the surgeon's stress levels, and contribute to DL-based anatomical landmarks identification in pancreatic surgery. Results In this study, the dataset was divided into training and testing sets, as shown in Table 1 . For the LDP group, 25 cases were included, resulting in 126 videos and 12,694 frames. The training set consisted of 10,434 frames from 19 patients, while the testing set included 2,260 frames from 6 patients. In the Whipple group, 30 cases were included, yielding 138 videos and a total of 35,986 frames. The training set comprised 28,915 frames from 22 patients, and the testing set included 7,071 frames from 8 patients (Supplementary Table 1). This distribution ensures a comprehensive evaluation of the model's performance across varying surgical contexts. Table 1 Data in training set and testing set Patient Video Total frame Training set Testing set LDP 25 126 12,694 10,434 2,260 Whipple 30 138 35,986 28,915 7,071 Recognition of SMV-PV axis and the splenic vein as venous anatomical landmarks Firstly, we tried individual recognition of SMV-PV axis and the splenic vein in LDP (Supplementary Table 1). Here we used multiclass segmentation to classify each image pixel into one of three categories: non-vein, SMV, or PV. To test the model performance in different image quality, we classified the surgery images into high and low difficulty groups (Table 2 , also see Methods and Supplementary Fig. 2). In the high difficulty group, recall ranged from 78.1–92.4%, while precision ranged from 71.0–85.5%. In the low difficulty group, recall ranged from 49.6–68.0%, and precision ranged from 66.3–79.5%. The highest mean Dice coefficient, 0.738, was achieved in the high difficulty group using the combined training set. However, we noticed that the multiclass segmentation recognition results were suboptimal (Supplementary Table 2), as the model might struggle to differentiate between specific types of blood vessels (see more in discussion). Anatomically, the splenic vein merges with the SMV to form the PV, so we hypothesized that treating them as a single entity could improve model performance. By merging the two veins into one unified "venous anatomical landmark," we aimed to enhance the model’s generalization ability and its applicability across different surgical procedures. As a result, the SMV-PV axis and splenic vein were classified as a single "combined vein" entity for all subsequent binary segmentation analyses of vein vs. non-vein. We then performed binary segmentation analyses of anatomical landmarks in LDP and Whipple surgeries (Table 2 ). In the LDP group, the model achieved a Recall of 84.10%, Precision of 76.30%, and Dice coefficient of 0.645 on low-difficulty cases, while performance on high-difficulty cases dropped to a Recall of 60.20%, Precision of 73.40%, and Dice of 0.465 (Table 2 ). In the Whipple group, results were similar, with a Recall of 82.50%, Precision of 86.60%, and Dice of 0.668 on low-difficulty cases, but lower performance on high-difficulty cases (Recall 61.90%, Precision 80.70%, Dice 0.512) (Table 2 ). Notably, using the combined dataset (All) yielded the highest performance on low-difficulty cases (Recall 92.40%, Precision 85.60%, Dice 0.738) and showed improved performance on high-difficulty cases (Recall 72.50%, Precision 80.50%, Dice 0.537) compared to either LDP or Whipple alone (Table 2 ). These results indicated strong model performance in the vein recognition. Moreover, our model, which requires only a A100 Tensor Core GPU, demonstrated strong real-time capability of 11 frames per second. Table 2 Binary Classification Results for Combined Vein Segmentation with Classified Difficulty Levels Training set Difficulty Combined vein testing set Recall Precision Dice LDP Low 84.10% 76.30% 0.645 High 60.20% 73.40% 0.465 Whipple Low 82.50% 86.60% 0.668 High 61.90% 80.70% 0.512 All Low 92.40% 85.60% 0.738 High 72.50% 80.50% 0.537 To further evaluate the model's generalization ability, we began by removing the boundary between difficulty levels. In LDP group, although the high/low difficulty ratio was nearly 2:1 (Supplementary Table 1), the overall recall for uniform difficulty levels closely matched that of the low-difficulty level. Interestingly, both precision and Dice scores improved when we combined the training sets. For example, the combined training set achieved a recall of 89.70% and precision of 93.50% on the LDP test set, resulting in a Dice score of 0.713 (Table 3 ). This suggests that combining different sub-training sets enhances the model's overall performance. Consequently, we merged the LDP and Whipple training sets and examined the test set under both individual and uniform conditions. Specifically, the Whipple test set with the combined training achieved a recall of 83.80%, precision of 90.40%, and a Dice score of 0.767, slightly outperforming the LDP test set's Dice score of 0.713 in the same combined training setup (Table 3 ). Additionally, when tested with both training and test sets combined, the model achieved intermediate performance, with a recall of 85.00%, precision of 91.10%, and Dice score of 0.754 (Table 3 ), sitting between the individual performances of LDP and Whipple. Furthermore, the combined dataset enables more accurate and comprehensive recognition. Compared to the model trained solely on the LDP dataset, the model trained on the combined dataset is capable of identifying previously unrecognized regions (identification) or further refining the recognition of existing regions (completion) on both LDP and Whipple test set (Fig. 3 ). Our model can accurately identify the SMV-PV axis in real-time during pancreatic surgery, enhancing precision and supporting surgeons with clear, reliable visualization throughout the procedure. We performed real-time processing on an 11-frame-per-second video (Supplementary Video 1) and non-real-time processing on a 24-frame-per-second video (Supplementary Video 2). In summary, integrating the SMV-PV axis and the splenic vein, along with unifying difficulty levels, resulted in combined training sets that enhanced the performance across different test sets. Table 3 Binary Classification Results for Combined Vein Segmentation with Uniform Difficulty Training Set Test Set Recall Precision False Negative False Positive Dice LDP LDP 83.60% 91.60% 16.40% 8.40% 0.686 Whipple Whipple 83.20% 89.20% 16.80% 10.80% 0.763 Combined LDP 89.70% 93.50% 10.30% 6.50% 0.713 Whipple 83.80% 90.40% 16.20% 9.60% 0.767 Combined 85.00% 91.10% 15.00% 8.90% 0.754 In summary, by integrating the veins, unifying difficulty levels, and combining training sets, our model demonstrated robust performance in the vein recognition task. The enhanced results achieved with the combined dataset indicate that the model generalizes well to different surgical contexts and can effectively support real-time vessel recognition, ensuring precision and safety in complex surgical operations. Discussion This study developed a machine learning approach for detecting the representative veins in laparoscopic pancreatic surgery, which is the first anatomical landmarks recognition for laparoscopic surgery to our best knowledge. We observed that dividing the training set into subsets initially yielded suboptimal results, whereas merging these subsets into a larger, more diverse dataset significantly enhanced performance. This highlights the importance of a high-capacity and diverse training set, reinforcing the well-established principle in machine learning that such datasets enable better generalization to new tasks 30 . Moreover, the model’s adaptability suggests that minimal fine-tuning could allow other researchers to effectively apply it to different datasets. After reviewing the processed videos, noticed that false positives were often caused by similar features, while the false negative cases were mainly caused by occlusion, followed by lack of distinction from surrounding tissues (Supplementary Fig. 4). Notably, we found that many small fragments were not eliminated during post-processing, leading to a high number of false positives. Adjusting the convolutional kernel from 7×7 to 69×69, resulting in a significant improvement in precision (Supplementary Fig. 4C, Supplementary Table 6), achieving the current performance. Furthermore, experienced surgeons prefer a clearer field of view during surgery, these small targets offer limited assistance, making our streamlined approach highly justified. However, when returning to multiclass segmentation task, the performance of our model was less satisfactory. Previous attempts by Nakamura 31 and Tokuyasu 32 , using the YoloV3 algorithm for multi-object detection during gallbladder resection without relying on pixel-level segmentation resulted in accuracy ranging only from 0.07 to 0.32. Despite subsequent enhancements to the algorithm elevating the mean Dice coefficient to 0.72, 0.49, 0.46, and 0.66, respectively, the results still felt short of expectations. Multiclass segmentation in anatomical landmarks involves categorizing different objects based on pixel-level segmentation. In this study, we initially carried out multiclass segmentation of the SMV-PV axis and splenic vein, two vital blood vessels in pancreatic surgery, obtaining a mean Dice coefficient of around 0.7. Based on those studies, we believe that multiclass segmentation is challenging now for the following reasons: 1. The indistinct visual contrast between those anatomical landmarks contributed to lower recognition accuracy, whereas the mean Dice coefficient for identification of gallbladder triangle is notably higher; 2. Obstructions of tissues or instruments often segmented the boundary of anatomic landmark structures, dividing them into several smaller segments, thereby complicating recognize; 3. Simultaneous objects in instance segmentation tasks may lead to a significant decrease in recognition accuracy 33 . Therefore, we applied binary segmentation for better model performance. Despite the overall progress medical imaging, there still remains a data scarcity across diverse medical scenarios 34 , 35 , wherein some certain investigations have been limited into merely a single case 36 , 37 . This data deficiency represents a critical bottleneck for the ongoing research in DL-powered medical applications. Our study aims to address this gap by supplementing data for venous anatomical landmarks in laparoscopic pancreatic surgery, thereby providing valuable resources to enhance the accuracy and safety of these procedures. With real-time processing capability, our model could have practical implications in surgical procedures. When combined with surgical robots, our model may assist surgeons in early identification and prevention of injuries by providing warnings and alerts before a mishap occurs, thereby enhancing the safety of the operation. Further improvements can be pursued based on the findings of this study. First, to enhance performance and improve the generalizability of this innovative model, it is crucial to gather more diverse and extensive datasets from a wide range of medical centers and surgeons. Such broader data collection would provide a richer foundation for training and refining the model, ensuring its adaptability to various surgical scenarios and techniques across different healthcare settings. Additionally, our current algorithm is unable to integrate contextual information for target recognition and maybe miss out on valuable insights. The limited perspective offered by a single image impedes the understanding of the surgical logic. Attempts to use models that account for contextual cues have been made, but their effectiveness remains less than ideal, for instance, achieving a mean Dice coefficient of 0.718 and a recall rate of 0.507 for renal artery identification 37 . Therefore, further development of the model is necessary to better leverage contextual information, enabling significant improvements in recognition accuracy. In summary, we conducted image segmentation in laparoscopic pancreatic surgery, identifying the contours of the SMV-PV axis and splenic vein, achieving commendable recognition results. Our study supplemented the research on automatic identification based on AI technology in pancreatic scenarios, confirming the effective identification of anatomical landmarks in pancreatic surgery. The model may further combine with laparoscopic and robotic surgical systems to enhance patients’ safety in pancreatic surgery. Methods 1. Dataset Generation Patients who underwent laparoscopic distal pancreatectomy at Peking Union Medical College Hospital between January 2021 and June 2022 were included. Inclusion criteria were: 1) Videos should contain the whole procedures of surgery, 2) Procedures should be laparoscopic distal pancreatectomy or Whipple, and 3) Appearance of the SMV-PV axis in the video for at least ≥ 2 minutes. Exclusion criteria were: 1) Any interruptions to laparoscopic surgery, and 2) Any history of abdominal or pelvic surgeries. All patients had signed the informed consent form for the collection of surgical videos, which was approved by the Ethics Committee of the hospital (SK-1901). To generate our data set, video data from those patients were randomly assigned in a 5:1 ratio to the training and testing sets. Then, one senior surgeon selected all videos with the SMV-PV axis and splenic veins and further extracted them at 1 frame per second to generate the image set. Four junior surgeons delineated the contours of the SMV-PV axis and splenic vein from these images, and classified them into high and low difficulty groups (Supplementary Fig. 2). Any image meeting either of the following criteria was classified into the high difficulty group: 1) The external rectangular area of the target ≤ 64×64 pixels; 2) Target vessels were unclear for any reasons such as blurred, poor exposure, overexposure, blood immersion, or fascial coverage (Supplementary Fig. 2). 2. Model Training All images were scaled down to 960×540 by a 4×4 convolutional kernel. Data augmentation, including random flipping, brightness enhancement, contrast enhancement, and saturation enhancement were performed. Subsequently, all pixels in those images underwent z-score normalization to minimize the noise 38 . The images were then fed into the High Resolution Network (HRNet)-Full Convolutional Network (FCN), which was pre-trained on ImageNet dataset, for training. HRNet maintained high resolution position-information throughout the whole process. There are two key features that could help HRNet to keep the information: 1. A parallel structure of feature maps with different resolutions; 2. An exchange of information between the high and low resolutions feature maps. In this study, HRNet is structured into 3 stages, and each stage comprises? 3 different steps: 1. Downsampling all existing feature maps to generate a new feature map with 1/2 resolution level (e.g. 1/4→1/8); 2. Applying convolution to all maps to extract new features; 3. Fusing information through upsampling and downsampling. Finally, all feature maps were fused and sent to FCN to restore them to original resolution and generated the predictions. The overall schematic of the HRNet_FCN module is illustrated in Fig. 1 . Please note that downsampling reduces the size of an image to decrease computation and capture essential features, while upsampling increases the image size to meet computational or network requirements. Due to the significantly lower total number of pixels in the SMV-PV axis and splenic vein (positive samples) compared to the number of background pixels (negative samples) at a ratio of approximately 1:15, the indispensable balance of positive and negative samples was adjusted by a cross-entropy loss function based on their weighted ratio. For a given image, the expression of the loss function was: $$\:\text{w}\text{e}\text{i}\text{g}\text{h}\text{t}\text{e}\text{d}\:\text{L}\text{o}\text{s}\text{s}=-\sum\:_{i}^{n}\frac{\left[\sum\:_{classes}{y}_{classweight}{y}_{true}\text{log}\left({y}_{pred}\right)\right]}{n}$$ where y true represented the label category, y pred represented the predicted label category of each pixel. n represented the total sum of all pixels in the image. The weight of each class (y classweight ) is generated as the reciprocal of the proportion of pixels in the entire image for that category. $$\:{\text{y}}_{classweight}=\frac{n}{{num\_y}_{class}}$$ Where num_y class represented the total number of pixels of different classes in ground truth. To eradicate spikes at the edges of the predicted areas and connect small predicted regions to achieve better recognition, morphological opening and closing were applied using the cv2.morphologyEx function from the Python library. The kernel size is 69×69 pixels. Training parameters were set as follows: batch size of 16, SGD optimizer, Softmax classifier, an initial learning rate of 0.01 with polynomial decay to a minimum of 0.0001, spanning 80,000 epochs. The model was trained on NVIDIA A100 Tensor Core GPU, with Python 3.6 and PyTorch 0.4.1. 3. Evaluation The Intersection over Union (IoU) was applied to assess the success predictions. When the IoU ratio of the predicted box and the ground truth box was ≥ a specific IoU threshold (in this study, usually 0.1 or 0.3), it was a true positive; otherwise, it was a false positive. Additionally, if a predicted box exists for a certain object but there is no corresponding ground truth box, it is called a false positive. Based on these results, precision and recall were calculated. The mean Dice coefficient was also used to evaluate the accuracy of model predictions. In terms of real-time performance, floating-point operations per second (FLOPs) were employed. The definitions of those parameters are in Supplement File1. Recall (sensitivity) refers to the proportion of ground truth boxes correctly detected by the model out of all ground truth boxes. Precision refers to the proportion of true positives among all predicted boxes. The formulas for both are as follows: Recall (Sensitivity) = \(\:\frac{\:\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}{\:\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}+\:\text{F}\text{a}\text{l}\text{s}\text{e}\:\text{N}\text{e}\text{g}\text{a}\text{t}\text{i}\text{v}\text{e}\text{s}}\) Precision = \(\:\frac{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}{\text{T}\text{r}\text{u}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}\:+\:\text{F}\text{a}\text{l}\text{s}\text{e}\:\text{P}\text{o}\text{s}\text{i}\text{t}\text{i}\text{v}\text{e}\text{s}}\) Dice coefficient is a metric to evaluate the model's detection performance. A higher mean Dice coefficient indicates better detection performance. The formula for calculating the mean Dice coefficient is as follows: mDice = \(\:\frac{2\:\text{*}\:\text{a}\text{r}\text{e}\text{a}\left(\text{P}\text{r}\text{e}\text{d}\text{i}\text{c}\text{t}\text{i}\text{o}\text{n}\right)\:\cap\:\:\text{a}\text{r}\text{e}\text{a}\left(\text{G}\text{r}\text{o}\text{u}\text{n}\text{d}\:\text{T}\text{r}\text{u}\text{t}\text{h}\right)}{\sum\:\text{a}\text{r}\text{e}\text{a}\left(\text{P}\text{r}\text{e}\text{d}\text{i}\text{c}\text{t}\text{i}\text{o}\text{n}\right)\:+\:\sum\:\text{a}\text{r}\text{e}\text{a}\left(\text{G}\text{r}\text{o}\text{u}\text{n}\text{d}\:\text{T}\text{r}\text{u}\text{t}\text{h}\right)}\) Declarations Data availability : The datasets used and analyzed during the current study available from the corresponding author on reasonable request. Code availability : The underlying code for this study is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author. Author Contributions: Data Collection, Z.W., J.G., C.F., X.H., R.C.; Parameter and Model Adjustment, Z.W., Q.Y.; Original Draft Preparation, Z.W., J.S., Q.Y., R.C., H.Z., S.H.; Review and Editing, Z.W., J.S., R.C., H.Z., S.H.; Supervision, S.H., W.W. Conflict of Interest: The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Disclosure of AI Assistance: This manuscript used AI tools for language editing, with all generated content reviewed and approved by the authors. The use of AI is transparently disclosed in accordance with journal guidelines. No patient data were processed using AI, ensuring full compliance with privacy and ethical standards Funding: This work was supported by National High Level Hospital Clinical Research Funding，No.2022-PUMCH-A-052. References Ammori, B. J. Pancreatic surgery in the laparoscopic era. Jop 4, 187–192 (2003). Nagakawa, Y. et al. The Straightened Splenic Vessels Method Improves Surgical Outcomes of Laparoscopic Distal Pancreatectomy. Dig Surg 34, 289–297 (2017). https://doi.org:10.1159/000452498 Liang, S., Hameed, U. & Jayaraman, S. Laparoscopic pancreatectomy: indications and outcomes. World J Gastroenterol 20, 14246–14254 (2014). https://doi.org:10.3748/wjg.v20.i39.14246 Kang, C. M. et al. Laparoscopic distal pancreatectomy with division of the pancreatic neck for benign and borderline malignant tumor in the proximal body of the pancreas. J Laparoendosc Adv Surg Tech A 20, 581–586 (2010). https://doi.org:10.1089/lap.2009.0348 Hellman, P. et al. Surgical strategy for large or malignant endocrine pancreatic tumors. World J Surg 24, 1353–1360 (2000). https://doi.org:10.1007/s002680010224 Pedrazzoli, S. Surgical Treatment of Pancreatic Cancer: Currently Debated Topics on Vascular Resection. Cancer Control 30, 10732748231153094 (2023). https://doi.org:10.1177/10732748231153094 Addeo, P. et al. Management of the splenic vein during a pancreaticoduodenectomy with venous resection for malignancy. Updates Surg 68, 241–246 (2016). https://doi.org:10.1007/s13304-016-0396-6 Bari, H., Wadhwani, S. & Dasari, B. V. M. Role of artificial intelligence in hepatobiliary and pancreatic surgery. World J Gastrointest Surg 13, 7–18 (2021). https://doi.org:10.4240/wjgs.v13.i1.7 Yao, Z. et al. Preoperative diagnosis and prediction of hepatocellular carcinoma: Radiomics analysis based on multi-modal ultrasound images. BMC Cancer 18, 1089 (2018). https://doi.org:10.1186/s12885-018-5003-4 van Helden, E. J. et al. Radiomics analysis of pre-treatment [(18)F]FDG PET/CT for patients with metastatic colorectal cancer undergoing palliative systemic treatment. Eur J Nucl Med Mol Imaging 45, 2307–2317 (2018). https://doi.org:10.1007/s00259-018-4100-6 Hosny, A., Parmar, C., Quackenbush, J., Schwartz, L. H. & Aerts, H. Artificial intelligence in radiology. Nat Rev Cancer 18, 500–510 (2018). https://doi.org:10.1038/s41568-018-0016-5 Hoang, U. N. et al. Assessment of multiphasic contrast-enhanced MR textures in differentiating small renal mass subtypes. Abdom Radiol (NY) 43, 3400–3409 (2018). https://doi.org:10.1007/s00261-018-1625-x Gulshan, V. et al. Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. Jama 316, 2402–2410 (2016). https://doi.org:10.1001/jama.2016.17216 Casella, A. et al. in 2020 25th International Conference on Pattern Recognition (ICPR). 6144–6149. Kitaguchi, D. et al. Real-time vascular anatomical image navigation for laparoscopic surgery: experimental study. Surg Endosc 36, 6105–6112 (2022). https://doi.org:10.1007/s00464-022-09384-7 Mascagni, P. et al. Artificial Intelligence for Surgical Safety: Automatic Assessment of the Critical View of Safety in Laparoscopic Cholecystectomy Using Deep Learning. Ann Surg 275, 955–961 (2022). https://doi.org:10.1097/sla.0000000000004351 Madani, A. et al. Artificial Intelligence for Intraoperative Guidance: Using Semantic Segmentation to Identify Surgical Anatomy During Laparoscopic Cholecystectomy. Annals of Surgery 276, 363–369 (2022). https://doi.org:10.1097/sla.0000000000004594 Wu, S. et al. SurgSmart: an artificial intelligent system for quality control in laparoscopic cholecystectomy: an observational study. Int J Surg 109, 1105–1114 (2023). https://doi.org:10.1097/js9.0000000000000329 Khan, D. Z. et al. Artificial intelligence assisted operative anatomy recognition in endoscopic pituitary surgery. npj Digital Medicine 7, 314 (2024). https://doi.org:10.1038/s41746-024-01273-8 Cheng, K. et al. Artificial intelligence-based automated laparoscopic cholecystectomy surgical phase recognition and analysis. Surg Endosc 36, 3160–3168 (2022). https://doi.org:10.1007/s00464-021-08619-3 Kudsi, O. Y., Gagner, M. & Jones, D. B. Laparoscopic distal pancreatectomy. Surg Oncol Clin N Am 22, 59–73, vi (2013). https://doi.org:10.1016/j.soc.2012.08.003 Chung, J. C., Kim, H. C. & Song, O. P. Laparoscopic distal pancreatectomy for benign or borderline malignant pancreatic tumors. Turk J Gastroenterol 25 Suppl 1, 162–166 (2014). https://doi.org:10.5152/tjg.2014.4389 Cai, H., Feng, L. & Peng, B. Laparoscopic pancreatectomy for benign or low-grade malignant pancreatic tumors: outcomes in a single high-volume institution. BMC Surgery 21, 412 (2021). https://doi.org:10.1186/s12893-021-01414-w Groot, V. P. et al. Patterns, Timing, and Predictors of Recurrence Following Pancreatectomy for Pancreatic Ductal Adenocarcinoma. Ann Surg 267, 936–945 (2018). https://doi.org:10.1097/sla.0000000000002234 Gagner, M. & Palermo, M. Laparoscopic Whipple procedure: review of the literature. J Hepatobiliary Pancreat Surg 16, 726–730 (2009). https://doi.org:10.1007/s00534-009-0142-2 Liao, C. H. et al. The feasibility of laparoscopic pancreaticoduodenectomy-a stepwise procedure and learning curve. Langenbecks Arch Surg 402, 853–861 (2017). https://doi.org:10.1007/s00423-016-1541-x Sahakyan, M. A. et al. Implementation and training with laparoscopic distal pancreatectomy: 23-year experience from a high-volume center. Surg Endosc 36, 468–479 (2022). https://doi.org:10.1007/s00464-021-08306-3 van Ramshorst, T. M. E. et al. Learning curves in laparoscopic distal pancreatectomy: a different experience for each generation. Int J Surg 109, 1648–1655 (2023). https://doi.org:10.1097/js9.0000000000000408 ke, S. et al. High-Resolution Representations for Labeling Pixels and Regions . (2019). Sun, C., Shrivastava, A., Singh, S. & Gupta, A. Revisiting Unreasonable Effectiveness of Data in Deep Learning Era. (2017). https://doi.org:10.48550/arXiv.1707.02968 Nakanuma, H. et al. An intraoperative artificial intelligence system identifying anatomical landmarks for laparoscopic cholecystectomy: a prospective clinical feasibility trial (J-SUMMIT-C-01). Surg Endosc 37, 1933–1942 (2023). https://doi.org:10.1007/s00464-022-09678-w Tokuyasu, T. et al. Development of an artificial intelligence system using deep learning to indicate anatomical landmarks during laparoscopic cholecystectomy. Surg Endosc 35, 1651–1658 (2021). https://doi.org:10.1007/s00464-020-07548-x Roß, T. et al. Comparative validation of multi-instance instrument segmentation in endoscopy: Results of the ROBUST-MIS 2019 challenge. Med Image Anal 70, 101920 (2021). https://doi.org:10.1016/j.media.2020.101920 Loukas, C., Gazis, A. & Schizas, D. Multiple instance convolutional neural network for gallbladder assessment from laparoscopic images. Int J Med Robot 18, e2445 (2022). https://doi.org:10.1002/rcs.2445 Leibetseder, A., Schoeffmann, K., Keckstein, J. & Keckstein, S. Post-surgical Endometriosis Segmentation in Laparoscopic Videos . (2021). Sonsilphong, S., Sonsilphong, A., Hormdee, D. & Khampitak, K. in 2022 International Electrical Engineering Congress (iEECON). 1–4. Caballas, K., Bolingot, H. J., Libatique, N. & Tangonan, G. Development of a Visual Guidance System for Laparoscopic Surgical Palpation using Computer Vision . (2021). Shorten, C. & Khoshgoftaar, T. M. A survey on Image Data Augmentation for Deep Learning. Journal of Big Data 6, 60 (2019). https://doi.org:10.1186/s40537-019-0197-0 Additional Declarations No competing interests reported. Supplementary Files Supplementarymatertial.docx supplementaryvideo1.mp4 supplementaryvideo2.mp4 Cite Share Download PDF Status: Published Journal Publication published 01 May, 2025 Read the published version in npj Digital Medicine → Version 1 posted Editorial decision: Revision requested 12 Mar, 2025 Reviews received at journal 18 Feb, 2025 Reviewers agreed at journal 15 Feb, 2025 Reviews received at journal 12 Dec, 2024 Reviewers agreed at journal 07 Dec, 2024 Reviewers agreed at journal 03 Dec, 2024 Reviewers agreed at journal 27 Nov, 2024 Reviewers invited by journal 24 Nov, 2024 Editor assigned by journal 19 Nov, 2024 Submission checks completed at journal 19 Nov, 2024 First submitted to journal 17 Nov, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5472618","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":387546874,"identity":"7a5c40d4-3de3-4394-8fc1-fc1088fa673a","order_by":0,"name":"Jile Shi","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Jile","middleName":"","lastName":"Shi","suffix":""},{"id":387546876,"identity":"089ed10e-6849-4bc3-a399-963d0b4ea4d8","order_by":1,"name":"Ruohan Cui","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Ruohan","middleName":"","lastName":"Cui","suffix":""},{"id":387546878,"identity":"57e199e5-8a26-4d01-af56-e218c8b68b53","order_by":2,"name":"Zhihong Wang","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Zhihong","middleName":"","lastName":"Wang","suffix":""},{"id":387546880,"identity":"2998ea64-efab-41f0-ad67-f7fee5f9c55e","order_by":3,"name":"Qi Yan","email":"","orcid":"","institution":"Tsinghua University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Qi","middleName":"","lastName":"Yan","suffix":""},{"id":387546881,"identity":"92f207e5-147c-407a-aa8d-31ea4fda3567","order_by":4,"name":"Lu Ping","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Lu","middleName":"","lastName":"Ping","suffix":""},{"id":387546883,"identity":"be9ef931-4fdd-4a12-98ec-6a7fd34e1885","order_by":5,"name":"Hu Zhou","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Hu","middleName":"","lastName":"Zhou","suffix":""},{"id":387546884,"identity":"cef74c0f-3d06-42f2-a765-e3d2c929ba5c","order_by":6,"name":"Junyi Gao","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Junyi","middleName":"","lastName":"Gao","suffix":""},{"id":387546885,"identity":"ffc6d5d6-9c82-4436-9db1-0380a6ee98b1","order_by":7,"name":"Chihua Fang","email":"","orcid":"","institution":"Zhujiang Hospital Southern Medical University","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Chihua","middleName":"","lastName":"Fang","suffix":""},{"id":387546886,"identity":"33825648-7343-4e77-be9f-3f5926c8b42e","order_by":8,"name":"Xianlin Han","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Xianlin","middleName":"","lastName":"Han","suffix":""},{"id":387546887,"identity":"238c2efc-bb5c-476a-bfcd-597d8c7598b9","order_by":9,"name":"Surong Hua","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABC0lEQVRIiWNgGAWjYBACAziLvYGBIQGMiNbCc5hkLRLJYIqwFnOJ5MefeffY5MlHvj/84eGO2jyD4wcYH1f8YpA3x6HFckaamTTPs7Riw9vJbBKJZ44XG5xJYDY828dguLMBh8NuJJgx8xw4nLhxdjIbQ2LbscQNBxLYJBt7GBIMDuDSkv75M1jLzMPMH8Bazj8gpCXHQBqkZb4EM4NEYltN4oYbQFsafuDRcuZNmeScA2mJG3iSzYBaDiTOvPGw2bCxQcJwAy4tx9M3f3hzwCZxfvvBxx9/ttUl9p1PPviw4Y+NPC5bEHohCoDxycDYwMDYJkFAPRDIN4CpOij3D2Edo2AUjIJRMGIAACvCaAOI3/A8AAAAAElFTkSuQmCC","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":true,"submittingAuthor":false,"prefix":"","firstName":"Surong","middleName":"","lastName":"Hua","suffix":""},{"id":387546889,"identity":"410ddc5e-09f6-4484-a2c6-66a5f9e78577","order_by":10,"name":"Wenming Wu","email":"","orcid":"","institution":"Chinese Academy of Medical Sciences \u0026 Peking Union Medical College","correspondingAuthor":false,"submittingAuthor":false,"prefix":"","firstName":"Wenming","middleName":"","lastName":"Wu","suffix":""}],"badges":[],"createdAt":"2024-11-18 04:23:30","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5472618/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5472618/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41746-025-01663-6","type":"published","date":"2025-05-01T15:56:58+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":70970100,"identity":"5755be2b-a913-407e-9cb3-822613ed6988","added_by":"auto","created_at":"2024-12-09 17:23:04","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":124439,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDiagrams of HRNet_FCN.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eA)\u003c/strong\u003eDiagram of High-Resolution Network (HRNet)-Full Convolutional Network (FCN). Images were scaled to 960×540 to input into HRNet_V2 Network for feature extraction. Feature maps at 1/4, 1/8, 1/16, and 1/32 levels were generated by HRNet_V2. After deconvolution, these feature maps were concatenated to form a fused feature map containing all information at different levels. Finally, the FCN performed convolution and deconvolution to obtain the result.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eB)\u003c/strong\u003eDetails of the parallel structure. The original image underwent three phases to form feature maps at four different scaling levels. Parallel arrows represented convolution, downward arrows represented downsampling, and upward arrows represented upsampling.\u003c/p\u003e","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/87b69a70435b45d8df488710.png"},{"id":70970102,"identity":"9d1efc06-18e6-481b-bfeb-c770878cff73","added_by":"auto","created_at":"2024-12-09 17:23:04","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":924457,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eResult of prediction. \u003c/strong\u003eThe blue box was the ground truth box, and the red box was the prediction box. The number was Intersection over Union (IoU) of prediction box and ground truth box.\u003c/p\u003e","description":"","filename":"floatimage23.png","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/e694713ec3295c3355df861e.png"},{"id":70970101,"identity":"4f89f8d2-d8d8-47ab-a50b-b7080d12a866","added_by":"auto","created_at":"2024-12-09 17:23:04","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":144264,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eComparison of the results of models trained solely on the LDP dataset and on the integrated dataset across different test sets. \u003c/strong\u003eThe red box represented the ground truth box, and the purple box represented the prediction box. The yellow arrows indicate regions identified by the model trained on the combined dataset that were missed by the model trained solely on the LDP dataset.\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/52dfefa1b901f26570262a0b.jpeg"},{"id":81988025,"identity":"6518dbc5-6fd4-420e-9244-dfb03d853079","added_by":"auto","created_at":"2025-05-05 16:07:27","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2015714,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/b5d0dcc7-43c7-4636-9a07-e8309eb4ce5c.pdf"},{"id":70970103,"identity":"21e65fa3-a52c-4dce-9b1a-28a113e03abc","added_by":"auto","created_at":"2024-12-09 17:23:05","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2766657,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymatertial.docx","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/80021bf74090a5561746d45a.docx"},{"id":70970802,"identity":"0f6e7802-5bd8-44cc-8feb-5367d3ca9902","added_by":"auto","created_at":"2024-12-09 17:31:05","extension":"mp4","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":18438347,"visible":true,"origin":"","legend":"","description":"","filename":"supplementaryvideo1.mp4","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/b7b394f2b371f0535f6d5440.mp4"},{"id":70970105,"identity":"c5d8a78a-237a-4a80-afda-7e4d807cd7f5","added_by":"auto","created_at":"2024-12-09 17:23:05","extension":"mp4","order_by":3,"title":"","display":"","copyAsset":false,"role":"supplement","size":18893497,"visible":true,"origin":"","legend":"","description":"","filename":"supplementaryvideo2.mp4","url":"https://assets-eu.researchsquare.com/files/rs-5472618/v1/93fed46078bac2315824091e.mp4"}],"financialInterests":"No competing interests reported.","formattedTitle":"Deep Learning HRNet-FCN for Blood Vessel Identification in Laparoscopic Pancreatic Surgery","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAs an increasingly popular treatment for pancreatic diseases, laparoscopic pancreatic surgery offers a minimally invasive approach, reducing recovery time, postoperative pain and hospital stays\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. However, laparoscopic pancreatic surgery remains highly challenging due to the complexity of the pancreas and surrounding structures. One of the main challenges for pancreatic surgery is the manipulation of critical anatomical vessels, especially the Superior Mesenteric Vein-Portal Vein (SMV-PV) axis and the splenic vein\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. These vessels pose a significant challenge due to their susceptibility to intraoperative bleeding, exacerbated by the veins\u0026rsquo; delicate nature\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. SMV-PV axis is a crucial factor in determining the resectability of pancreatic tumors, especially when venous invasion occurs, requiring preservation or reconstruction for successful surgery \u003csup\u003e\u003cspan additionalcitationids=\"CR5\" citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. Proper management of this axis during venous resection is essential to minimize complications and maintain splenic vein function, which significantly impacts surgical outcomes\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Therefore, accurate identification and careful handling of the SMV-PV axis and splenic vein is one of the most difficult complications in pancreatic surgery\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. However, as the surgeon operating with the laparoscopy cannot use the \"sense of touch\" to identify blood\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e, it is important to enhance the visual identification of SMV-PV axis and splenic vein during laparoscopic pancreatic surgery.\u003c/p\u003e \u003cp\u003eIn the field of medical imaging recognition, Deep Learning (DL) technology has catalyzed significant breakthroughs in previous medical imaging studies, like ultrasound\u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, PET-CT\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, CT\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e, MR\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e and Retinal Fundus Photographs\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. Specifically, in the realm of intelligent surgery, DL has proven its merit in accurately identifying and segmenting critical arteries, such as renal\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e and mesenteric arteries\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e, achieving impressive precision of 0.937. Over the past three years, several leading publications have reported the application of DL technology in identifying anatomical landmarks and safety assessments in surgical procedures\u003csup\u003e\u003cspan additionalcitationids=\"CR17 CR18 CR19\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. However, these studies predominantly focused on cholecystectomy and endoscopic pituitary surgery, leaving other surgical scenarios, such as pancreatic surgery, relatively unexplored.\u003c/p\u003e \u003cp\u003eHere, we present two examples of SMV-PV axis and the splenic vein recognition in laparoscopic pancreatic surgery, including Laparoscopic Distal Pancreatectomy (LDP) and Pancreaticoduodenectomy (Whipple procedure). LDP and Whipple are both widely recognized as a standard procedure for both benign and malignant pancreatic diseases. LDP is particularly for lesions located in the pancreatic tail or body\u003csup\u003e\u003cspan additionalcitationids=\"CR22 CR23\" citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e, while the Whipple procedure is primarily performed for lesions located in the pancreatic head, as well as the duodenum, common bile duct, or surrounding areas\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. However, both LDP and Whipple are complex surgical procedure that requires high level qualifications and trainings for surgeons: surgeons typically need to complete hundreds of laparoscopic procedures in other areas under to develop the expertise necessary for safe performance of LDP\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e and Whipple \u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e,\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e. Therefore, it is important to enhance the venous anatomical landmarks for LDP Whipple procedure, for assisting surgery and educational purpose.\u003c/p\u003e \u003cp\u003eIn both processes, we constructed an annotated database of SMV-PV images from experienced surgeons. We then employed the High Resolution Network (HRNet) \u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e to train the model, which identified and delineated the SMV-PV axis and splenic vein in LDP, and explored the real time segmentation of anatomical landmarks. Our model allows for instant identification of the SMV-PV axis during pancreatic surgery, enhancing surgical precision. Overall, this work can enhance the safety, reduces the surgeon's stress levels, and contribute to DL-based anatomical landmarks identification in pancreatic surgery.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eIn this study, the dataset was divided into training and testing sets, as shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e. For the LDP group, 25 cases were included, resulting in 126 videos and 12,694 frames. The training set consisted of 10,434 frames from 19 patients, while the testing set included 2,260 frames from 6 patients. In the Whipple group, 30 cases were included, yielding 138 videos and a total of 35,986 frames. The training set comprised 28,915 frames from 22 patients, and the testing set included 7,071 frames from 8 patients (Supplementary Table\u0026nbsp;1). This distribution ensures a comprehensive evaluation of the model's performance across varying surgical contexts.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eData in training set and testing set\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatient\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eVideo\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTotal frame\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eTesting set\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e25\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e126\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e12,694\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10,434\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e2,260\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhipple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e138\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e35,986\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e28,915\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e7,071\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eRecognition of SMV-PV axis and the splenic vein as venous anatomical landmarks\u003c/h2\u003e \u003cp\u003eFirstly, we tried individual recognition of SMV-PV axis and the splenic vein in LDP (Supplementary Table\u0026nbsp;1). Here we used multiclass segmentation to classify each image pixel into one of three categories: non-vein, SMV, or PV. To test the model performance in different image quality, we classified the surgery images into high and low difficulty groups (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, also see Methods and Supplementary Fig.\u0026nbsp;2). In the high difficulty group, recall ranged from 78.1\u0026ndash;92.4%, while precision ranged from 71.0\u0026ndash;85.5%. In the low difficulty group, recall ranged from 49.6\u0026ndash;68.0%, and precision ranged from 66.3\u0026ndash;79.5%. The highest mean Dice coefficient, 0.738, was achieved in the high difficulty group using the combined training set.\u003c/p\u003e \u003cp\u003eHowever, we noticed that the multiclass segmentation recognition results were suboptimal (Supplementary Table\u0026nbsp;2), as the model might struggle to differentiate between specific types of blood vessels (see more in discussion). Anatomically, the splenic vein merges with the SMV to form the PV, so we hypothesized that treating them as a single entity could improve model performance. By merging the two veins into one unified \"venous anatomical landmark,\" we aimed to enhance the model\u0026rsquo;s generalization ability and its applicability across different surgical procedures. As a result, the SMV-PV axis and splenic vein were classified as a single \"combined vein\" entity for all subsequent binary segmentation analyses of vein vs. non-vein.\u003c/p\u003e \u003cp\u003eWe then performed binary segmentation analyses of anatomical landmarks in LDP and Whipple surgeries (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In the LDP group, the model achieved a Recall of 84.10%, Precision of 76.30%, and Dice coefficient of 0.645 on low-difficulty cases, while performance on high-difficulty cases dropped to a Recall of 60.20%, Precision of 73.40%, and Dice of 0.465 (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). In the Whipple group, results were similar, with a Recall of 82.50%, Precision of 86.60%, and Dice of 0.668 on low-difficulty cases, but lower performance on high-difficulty cases (Recall 61.90%, Precision 80.70%, Dice 0.512) (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Notably, using the combined dataset (All) yielded the highest performance on low-difficulty cases (Recall 92.40%, Precision 85.60%, Dice 0.738) and showed improved performance on high-difficulty cases (Recall 72.50%, Precision 80.50%, Dice 0.537) compared to either LDP or Whipple alone (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). These results indicated strong model performance in the vein recognition. Moreover, our model, which requires only a A100 Tensor Core GPU, demonstrated strong real-time capability of 11 frames per second.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBinary Classification Results for Combined Vein Segmentation with Classified Difficulty Levels\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eDifficulty\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"3\" nameend=\"c5\" namest=\"c3\"\u003e \u003cp\u003eCombined vein testing set\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDice\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eLDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e84.10%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e76.30%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.645\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e60.20%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e73.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.465\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eWhipple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e82.50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e86.60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.668\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e61.90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80.70%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.512\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAll\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLow\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e92.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e85.60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.738\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eHigh\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e72.50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e80.50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.537\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTo further evaluate the model's generalization ability, we began by removing the boundary between difficulty levels. In LDP group, although the high/low difficulty ratio was nearly 2:1 (Supplementary Table\u0026nbsp;1), the overall recall for uniform difficulty levels closely matched that of the low-difficulty level. Interestingly, both precision and Dice scores improved when we combined the training sets. For example, the combined training set achieved a recall of 89.70% and precision of 93.50% on the LDP test set, resulting in a Dice score of 0.713 (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). This suggests that combining different sub-training sets enhances the model's overall performance. Consequently, we merged the LDP and Whipple training sets and examined the test set under both individual and uniform conditions. Specifically, the Whipple test set with the combined training achieved a recall of 83.80%, precision of 90.40%, and a Dice score of 0.767, slightly outperforming the LDP test set's Dice score of 0.713 in the same combined training setup (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Additionally, when tested with both training and test sets combined, the model achieved intermediate performance, with a recall of 85.00%, precision of 91.10%, and Dice score of 0.754 (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e), sitting between the individual performances of LDP and Whipple.\u003c/p\u003e \u003cp\u003eFurthermore, the combined dataset enables more accurate and comprehensive recognition. Compared to the model trained solely on the LDP dataset, the model trained on the combined dataset is capable of identifying previously unrecognized regions (identification) or further refining the recognition of existing regions (completion) on both LDP and Whipple test set (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Our model can accurately identify the SMV-PV axis in real-time during pancreatic surgery, enhancing precision and supporting surgeons with clear, reliable visualization throughout the procedure. We performed real-time processing on an 11-frame-per-second video (Supplementary Video 1) and non-real-time processing on a 24-frame-per-second video (Supplementary Video 2).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eIn summary, integrating the SMV-PV axis and the splenic vein, along with unifying difficulty levels, resulted in combined training sets that enhanced the performance across different test sets.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eBinary Classification Results for Combined Vein Segmentation with Uniform Difficulty\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTraining Set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest Set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eFalse Negative\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eFalse Positive\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eDice\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e91.60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.686\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWhipple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWhipple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.20%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e89.20%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.80%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e10.80%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.763\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eCombined\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eLDP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e89.70%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e93.50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e10.30%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e6.50%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.713\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eWhipple\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e83.80%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e90.40%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e16.20%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e9.60%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.767\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eCombined\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e85.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e91.10%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e15.00%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e8.90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.754\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eIn summary, by integrating the veins, unifying difficulty levels, and combining training sets, our model demonstrated robust performance in the vein recognition task. The enhanced results achieved with the combined dataset indicate that the model generalizes well to different surgical contexts and can effectively support real-time vessel recognition, ensuring precision and safety in complex surgical operations.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study developed a machine learning approach for detecting the representative veins in laparoscopic pancreatic surgery, which is the first anatomical landmarks recognition for laparoscopic surgery to our best knowledge. We observed that dividing the training set into subsets initially yielded suboptimal results, whereas merging these subsets into a larger, more diverse dataset significantly enhanced performance. This highlights the importance of a high-capacity and diverse training set, reinforcing the well-established principle in machine learning that such datasets enable better generalization to new tasks\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e. Moreover, the model\u0026rsquo;s adaptability suggests that minimal fine-tuning could allow other researchers to effectively apply it to different datasets.\u003c/p\u003e \u003cp\u003eAfter reviewing the processed videos, noticed that false positives were often caused by similar features, while the false negative cases were mainly caused by occlusion, followed by lack of distinction from surrounding tissues (Supplementary Fig.\u0026nbsp;4). Notably, we found that many small fragments were not eliminated during post-processing, leading to a high number of false positives. Adjusting the convolutional kernel from 7\u0026times;7 to 69\u0026times;69, resulting in a significant improvement in precision (Supplementary Fig.\u0026nbsp;4C, Supplementary Table\u0026nbsp;6), achieving the current performance. Furthermore, experienced surgeons prefer a clearer field of view during surgery, these small targets offer limited assistance, making our streamlined approach highly justified.\u003c/p\u003e \u003cp\u003eHowever, when returning to multiclass segmentation task, the performance of our model was less satisfactory. Previous attempts by Nakamura \u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e\u003c/sup\u003e and Tokuyasu \u003csup\u003e\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e, using the YoloV3 algorithm for multi-object detection during gallbladder resection without relying on pixel-level segmentation resulted in accuracy ranging only from 0.07 to 0.32. Despite subsequent enhancements to the algorithm elevating the mean Dice coefficient to 0.72, 0.49, 0.46, and 0.66, respectively, the results still felt short of expectations. Multiclass segmentation in anatomical landmarks involves categorizing different objects based on pixel-level segmentation. In this study, we initially carried out multiclass segmentation of the SMV-PV axis and splenic vein, two vital blood vessels in pancreatic surgery, obtaining a mean Dice coefficient of around 0.7. Based on those studies, we believe that multiclass segmentation is challenging now for the following reasons: 1. The indistinct visual contrast between those anatomical landmarks contributed to lower recognition accuracy, whereas the mean Dice coefficient for identification of gallbladder triangle is notably higher; 2. Obstructions of tissues or instruments often segmented the boundary of anatomic landmark structures, dividing them into several smaller segments, thereby complicating recognize; 3. Simultaneous objects in instance segmentation tasks may lead to a significant decrease in recognition accuracy \u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e. Therefore, we applied binary segmentation for better model performance.\u003c/p\u003e \u003cp\u003eDespite the overall progress medical imaging, there still remains a data scarcity across diverse medical scenarios \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e,\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e, wherein some certain investigations have been limited into merely a single case \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e,\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. This data deficiency represents a critical bottleneck for the ongoing research in DL-powered medical applications. Our study aims to address this gap by supplementing data for venous anatomical landmarks in laparoscopic pancreatic surgery, thereby providing valuable resources to enhance the accuracy and safety of these procedures. With real-time processing capability, our model could have practical implications in surgical procedures. When combined with surgical robots, our model may assist surgeons in early identification and prevention of injuries by providing warnings and alerts before a mishap occurs, thereby enhancing the safety of the operation.\u003c/p\u003e \u003cp\u003eFurther improvements can be pursued based on the findings of this study. First, to enhance performance and improve the generalizability of this innovative model, it is crucial to gather more diverse and extensive datasets from a wide range of medical centers and surgeons. Such broader data collection would provide a richer foundation for training and refining the model, ensuring its adaptability to various surgical scenarios and techniques across different healthcare settings. Additionally, our current algorithm is unable to integrate contextual information for target recognition and maybe miss out on valuable insights. The limited perspective offered by a single image impedes the understanding of the surgical logic. Attempts to use models that account for contextual cues have been made, but their effectiveness remains less than ideal, for instance, achieving a mean Dice coefficient of 0.718 and a recall rate of 0.507 for renal artery identification\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e. Therefore, further development of the model is necessary to better leverage contextual information, enabling significant improvements in recognition accuracy.\u003c/p\u003e \u003cp\u003eIn summary, we conducted image segmentation in laparoscopic pancreatic surgery, identifying the contours of the SMV-PV axis and splenic vein, achieving commendable recognition results. Our study supplemented the research on automatic identification based on AI technology in pancreatic scenarios, confirming the effective identification of anatomical landmarks in pancreatic surgery. The model may further combine with laparoscopic and robotic surgical systems to enhance patients\u0026rsquo; safety in pancreatic surgery.\u003c/p\u003e "},{"header":"Methods","content":"\n\u003ch3\u003e1. Dataset Generation\u003c/h3\u003e\n\u003cp\u003ePatients who underwent laparoscopic distal pancreatectomy at Peking Union Medical College Hospital between January 2021 and June 2022 were included. Inclusion criteria were: 1) Videos should contain the whole procedures of surgery, 2) Procedures should be laparoscopic distal pancreatectomy or Whipple, and 3) Appearance of the SMV-PV axis in the video for at least\u0026thinsp;\u0026ge;\u0026thinsp;2 minutes. Exclusion criteria were: 1) Any interruptions to laparoscopic surgery, and 2) Any history of abdominal or pelvic surgeries. All patients had signed the informed consent form for the collection of surgical videos, which was approved by the Ethics Committee of the hospital (SK-1901).\u003c/p\u003e \u003cp\u003eTo generate our data set, video data from those patients were randomly assigned in a 5:1 ratio to the training and testing sets. Then, one senior surgeon selected all videos with the SMV-PV axis and splenic veins and further extracted them at 1 frame per second to generate the image set. Four junior surgeons delineated the contours of the SMV-PV axis and splenic vein from these images, and classified them into high and low difficulty groups (Supplementary Fig.\u0026nbsp;2). Any image meeting either of the following criteria was classified into the high difficulty group: 1) The external rectangular area of the target\u0026thinsp;\u0026le;\u0026thinsp;64\u0026times;64 pixels; 2) Target vessels were unclear for any reasons such as blurred, poor exposure, overexposure, blood immersion, or fascial coverage (Supplementary Fig.\u0026nbsp;2).\u003c/p\u003e\n\u003ch3\u003e2. Model Training\u003c/h3\u003e\n\u003cp\u003eAll images were scaled down to 960\u0026times;540 by a 4\u0026times;4 convolutional kernel. Data augmentation, including random flipping, brightness enhancement, contrast enhancement, and saturation enhancement were performed. Subsequently, all pixels in those images underwent z-score normalization to minimize the noise\u003csup\u003e\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u003c/sup\u003e. The images were then fed into the High Resolution Network (HRNet)-Full Convolutional Network (FCN), which was pre-trained on ImageNet dataset, for training.\u003c/p\u003e \u003cp\u003eHRNet maintained high resolution position-information throughout the whole process. There are two key features that could help HRNet to keep the information: 1. A parallel structure of feature maps with different resolutions; 2. An exchange of information between the high and low resolutions feature maps. In this study, HRNet is structured into 3 stages, and each stage comprises? 3 different steps: 1. Downsampling all existing feature maps to generate a new feature map with 1/2 resolution level (e.g. 1/4\u0026rarr;1/8); 2. Applying convolution to all maps to extract new features; 3. Fusing information through upsampling and downsampling. Finally, all feature maps were fused and sent to FCN to restore them to original resolution and generated the predictions. The overall schematic of the HRNet_FCN module is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e1\u003c/span\u003e. Please note that downsampling reduces the size of an image to decrease computation and capture essential features, while upsampling increases the image size to meet computational or network requirements.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eDue to the significantly lower total number of pixels in the SMV-PV axis and splenic vein (positive samples) compared to the number of background pixels (negative samples) at a ratio of approximately 1:15, the indispensable balance of positive and negative samples was adjusted by a cross-entropy loss function based on their weighted ratio. For a given image, the expression of the loss function was:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:\\text{w}\\text{e}\\text{i}\\text{g}\\text{h}\\text{t}\\text{e}\\text{d}\\:\\text{L}\\text{o}\\text{s}\\text{s}=-\\sum\\:_{i}^{n}\\frac{\\left[\\sum\\:_{classes}{y}_{classweight}{y}_{true}\\text{log}\\left({y}_{pred}\\right)\\right]}{n}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003ewhere y\u003csub\u003etrue\u003c/sub\u003e represented the label category, y\u003csub\u003epred\u003c/sub\u003e represented the predicted label category of each pixel. n represented the total sum of all pixels in the image. The weight of each class (y\u003csub\u003eclassweight\u003c/sub\u003e) is generated as the reciprocal of the proportion of pixels in the entire image for that category.\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:{\\text{y}}_{classweight}=\\frac{n}{{num\\_y}_{class}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eWhere num_y\u003csub\u003eclass\u003c/sub\u003e represented the total number of pixels of different classes in ground truth.\u003c/p\u003e \u003cp\u003eTo eradicate spikes at the edges of the predicted areas and connect small predicted regions to achieve better recognition, morphological opening and closing were applied using the cv2.morphologyEx function from the Python library. The kernel size is 69\u0026times;69 pixels.\u003c/p\u003e \u003cp\u003eTraining parameters were set as follows: batch size of 16, SGD optimizer, Softmax classifier, an initial learning rate of 0.01 with polynomial decay to a minimum of 0.0001, spanning 80,000 epochs. The model was trained on NVIDIA A100 Tensor Core GPU, with Python 3.6 and PyTorch 0.4.1.\u003c/p\u003e\n\u003ch3\u003e3. Evaluation\u003c/h3\u003e\n\u003cp\u003eThe Intersection over Union (IoU) was applied to assess the success predictions. When the IoU ratio of the predicted box and the ground truth box was \u0026ge;\u0026thinsp;a specific IoU threshold (in this study, usually 0.1 or 0.3), it was a true positive; otherwise, it was a false positive. Additionally, if a predicted box exists for a certain object but there is no corresponding ground truth box, it is called a false positive. Based on these results, precision and recall were calculated. The mean Dice coefficient was also used to evaluate the accuracy of model predictions. In terms of real-time performance, floating-point operations per second (FLOPs) were employed. The definitions of those parameters are in Supplement File1.\u003c/p\u003e \u003cp\u003eRecall (sensitivity) refers to the proportion of ground truth boxes correctly detected by the model out of all ground truth boxes. Precision refers to the proportion of true positives among all predicted boxes. The formulas for both are as follows:\u003c/p\u003e \u003cp\u003eRecall (Sensitivity) =\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{\\:\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}{\\:\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}+\\:\\text{F}\\text{a}\\text{l}\\text{s}\\text{e}\\:\\text{N}\\text{e}\\text{g}\\text{a}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003cp\u003ePrecision =\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}{\\text{T}\\text{r}\\text{u}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}\\:+\\:\\text{F}\\text{a}\\text{l}\\text{s}\\text{e}\\:\\text{P}\\text{o}\\text{s}\\text{i}\\text{t}\\text{i}\\text{v}\\text{e}\\text{s}}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e \u003cp\u003eDice coefficient is a metric to evaluate the model's detection performance. A higher mean Dice coefficient indicates better detection performance. The formula for calculating the mean Dice coefficient is as follows:\u003c/p\u003e \u003cp\u003emDice = \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\frac{2\\:\\text{*}\\:\\text{a}\\text{r}\\text{e}\\text{a}\\left(\\text{P}\\text{r}\\text{e}\\text{d}\\text{i}\\text{c}\\text{t}\\text{i}\\text{o}\\text{n}\\right)\\:\\cap\\:\\:\\text{a}\\text{r}\\text{e}\\text{a}\\left(\\text{G}\\text{r}\\text{o}\\text{u}\\text{n}\\text{d}\\:\\text{T}\\text{r}\\text{u}\\text{t}\\text{h}\\right)}{\\sum\\:\\text{a}\\text{r}\\text{e}\\text{a}\\left(\\text{P}\\text{r}\\text{e}\\text{d}\\text{i}\\text{c}\\text{t}\\text{i}\\text{o}\\text{n}\\right)\\:+\\:\\sum\\:\\text{a}\\text{r}\\text{e}\\text{a}\\left(\\text{G}\\text{r}\\text{o}\\text{u}\\text{n}\\text{d}\\:\\text{T}\\text{r}\\text{u}\\text{t}\\text{h}\\right)}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e: The datasets used and analyzed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e: The underlying code for this study is not publicly available but may be made available to qualified researchers on reasonable request from the corresponding author.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor Contributions:\u003c/strong\u003e Data Collection, Z.W., J.G., C.F., X.H., R.C.; Parameter and Model Adjustment, Z.W., Q.Y.; Original Draft Preparation, Z.W., J.S., Q.Y., R.C., H.Z., S.H.; Review and Editing, Z.W., J.S., R.C., H.Z., S.H.; Supervision, S.H., W.W.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflict of Interest:\u003c/strong\u003e The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDisclosure of AI Assistance:\u003c/strong\u003e This manuscript used AI tools for language editing, with all generated content reviewed and approved by the authors. The use of AI is transparently disclosed in accordance with journal guidelines. No patient data were processed using AI, ensuring full compliance with privacy and ethical standards\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding: \u003c/strong\u003eThis work was supported\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eby\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003eNational High Level Hospital Clinical Research Funding，No.2022-PUMCH-A-052.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAmmori, B. J. Pancreatic surgery in the laparoscopic era. Jop 4, 187\u0026ndash;192 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNagakawa, Y. \u003cem\u003eet al.\u003c/em\u003e The Straightened Splenic Vessels Method Improves Surgical Outcomes of Laparoscopic Distal Pancreatectomy. Dig Surg 34, 289\u0026ndash;297 (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1159/000452498\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1159/000452498\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiang, S., Hameed, U. \u0026amp; Jayaraman, S. Laparoscopic pancreatectomy: indications and outcomes. World J Gastroenterol 20, 14246\u0026ndash;14254 (2014). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.3748/wjg.v20.i39.14246\u003c/span\u003e\u003cspan address=\"https://doi.org:10.3748/wjg.v20.i39.14246\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKang, C. M. \u003cem\u003eet al.\u003c/em\u003e Laparoscopic distal pancreatectomy with division of the pancreatic neck for benign and borderline malignant tumor in the proximal body of the pancreas. J Laparoendosc Adv Surg Tech A 20, 581\u0026ndash;586 (2010). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1089/lap.2009.0348\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1089/lap.2009.0348\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHellman, P. \u003cem\u003eet al.\u003c/em\u003e Surgical strategy for large or malignant endocrine pancreatic tumors. World J Surg 24, 1353\u0026ndash;1360 (2000). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s002680010224\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s002680010224\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePedrazzoli, S. Surgical Treatment of Pancreatic Cancer: Currently Debated Topics on Vascular Resection. Cancer Control 30, 10732748231153094 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1177/10732748231153094\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1177/10732748231153094\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAddeo, P. \u003cem\u003eet al.\u003c/em\u003e Management of the splenic vein during a pancreaticoduodenectomy with venous resection for malignancy. Updates Surg 68, 241\u0026ndash;246 (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s13304-016-0396-6\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s13304-016-0396-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBari, H., Wadhwani, S. \u0026amp; Dasari, B. V. M. Role of artificial intelligence in hepatobiliary and pancreatic surgery. World J Gastrointest Surg 13, 7\u0026ndash;18 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.4240/wjgs.v13.i1.7\u003c/span\u003e\u003cspan address=\"https://doi.org:10.4240/wjgs.v13.i1.7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYao, Z. \u003cem\u003eet al.\u003c/em\u003e Preoperative diagnosis and prediction of hepatocellular carcinoma: Radiomics analysis based on multi-modal ultrasound images. BMC Cancer 18, 1089 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1186/s12885-018-5003-4\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1186/s12885-018-5003-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Helden, E. J. \u003cem\u003eet al.\u003c/em\u003e Radiomics analysis of pre-treatment [(18)F]FDG PET/CT for patients with metastatic colorectal cancer undergoing palliative systemic treatment. Eur J Nucl Med Mol Imaging 45, 2307\u0026ndash;2317 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00259-018-4100-6\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00259-018-4100-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHosny, A., Parmar, C., Quackenbush, J., Schwartz, L. H. \u0026amp; Aerts, H. Artificial intelligence in radiology. Nat Rev Cancer 18, 500\u0026ndash;510 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1038/s41568-018-0016-5\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1038/s41568-018-0016-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoang, U. N. \u003cem\u003eet al.\u003c/em\u003e Assessment of multiphasic contrast-enhanced MR textures in differentiating small renal mass subtypes. Abdom Radiol (NY) 43, 3400\u0026ndash;3409 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00261-018-1625-x\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00261-018-1625-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGulshan, V. \u003cem\u003eet al.\u003c/em\u003e Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs. Jama 316, 2402\u0026ndash;2410 (2016). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1001/jama.2016.17216\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1001/jama.2016.17216\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCasella, A. \u003cem\u003eet al.\u003c/em\u003e in \u003cem\u003e2020 25th International Conference on Pattern Recognition (ICPR).\u003c/em\u003e 6144\u0026ndash;6149.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKitaguchi, D. \u003cem\u003eet al.\u003c/em\u003e Real-time vascular anatomical image navigation for laparoscopic surgery: experimental study. Surg Endosc 36, 6105\u0026ndash;6112 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00464-022-09384-7\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00464-022-09384-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMascagni, P. \u003cem\u003eet al.\u003c/em\u003e Artificial Intelligence for Surgical Safety: Automatic Assessment of the Critical View of Safety in Laparoscopic Cholecystectomy Using Deep Learning. Ann Surg 275, 955\u0026ndash;961 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1097/sla.0000000000004351\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1097/sla.0000000000004351\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMadani, A. \u003cem\u003eet al.\u003c/em\u003e Artificial Intelligence for Intraoperative Guidance: Using Semantic Segmentation to Identify Surgical Anatomy During Laparoscopic Cholecystectomy. Annals of Surgery 276, 363\u0026ndash;369 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1097/sla.0000000000004594\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1097/sla.0000000000004594\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu, S. \u003cem\u003eet al.\u003c/em\u003e SurgSmart: an artificial intelligent system for quality control in laparoscopic cholecystectomy: an observational study. Int J Surg 109, 1105\u0026ndash;1114 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1097/js9.0000000000000329\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1097/js9.0000000000000329\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKhan, D. Z. \u003cem\u003eet al.\u003c/em\u003e Artificial intelligence assisted operative anatomy recognition in endoscopic pituitary surgery. npj Digital Medicine 7, 314 (2024). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1038/s41746-024-01273-8\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1038/s41746-024-01273-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng, K. \u003cem\u003eet al.\u003c/em\u003e Artificial intelligence-based automated laparoscopic cholecystectomy surgical phase recognition and analysis. Surg Endosc 36, 3160\u0026ndash;3168 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00464-021-08619-3\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00464-021-08619-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKudsi, O. Y., Gagner, M. \u0026amp; Jones, D. B. Laparoscopic distal pancreatectomy. \u003cem\u003eSurg Oncol Clin N Am\u003c/em\u003e 22, 59\u0026ndash;73, vi (2013). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1016/j.soc.2012.08.003\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1016/j.soc.2012.08.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChung, J. C., Kim, H. C. \u0026amp; Song, O. P. Laparoscopic distal pancreatectomy for benign or borderline malignant pancreatic tumors. Turk J Gastroenterol 25 Suppl 1, 162\u0026ndash;166 (2014). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.5152/tjg.2014.4389\u003c/span\u003e\u003cspan address=\"https://doi.org:10.5152/tjg.2014.4389\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCai, H., Feng, L. \u0026amp; Peng, B. Laparoscopic pancreatectomy for benign or low-grade malignant pancreatic tumors: outcomes in a single high-volume institution. BMC Surgery 21, 412 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1186/s12893-021-01414-w\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1186/s12893-021-01414-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGroot, V. P. \u003cem\u003eet al.\u003c/em\u003e Patterns, Timing, and Predictors of Recurrence Following Pancreatectomy for Pancreatic Ductal Adenocarcinoma. Ann Surg 267, 936\u0026ndash;945 (2018). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1097/sla.0000000000002234\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1097/sla.0000000000002234\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGagner, M. \u0026amp; Palermo, M. Laparoscopic Whipple procedure: review of the literature. J Hepatobiliary Pancreat Surg 16, 726\u0026ndash;730 (2009). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00534-009-0142-2\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00534-009-0142-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiao, C. H. \u003cem\u003eet al.\u003c/em\u003e The feasibility of laparoscopic pancreaticoduodenectomy-a stepwise procedure and learning curve. Langenbecks Arch Surg 402, 853\u0026ndash;861 (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00423-016-1541-x\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00423-016-1541-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSahakyan, M. A. \u003cem\u003eet al.\u003c/em\u003e Implementation and training with laparoscopic distal pancreatectomy: 23-year experience from a high-volume center. Surg Endosc 36, 468\u0026ndash;479 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00464-021-08306-3\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00464-021-08306-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan Ramshorst, T. M. E. \u003cem\u003eet al.\u003c/em\u003e Learning curves in laparoscopic distal pancreatectomy: a different experience for each generation. Int J Surg 109, 1648\u0026ndash;1655 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1097/js9.0000000000000408\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1097/js9.0000000000000408\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eke, S. \u003cem\u003eet al. High-Resolution Representations for Labeling Pixels and Regions\u003c/em\u003e. (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun, C., Shrivastava, A., Singh, S. \u0026amp; Gupta, A. Revisiting Unreasonable Effectiveness of Data in Deep Learning Era. (2017). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.48550/arXiv.1707.02968\u003c/span\u003e\u003cspan address=\"https://doi.org:10.48550/arXiv.1707.02968\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNakanuma, H. \u003cem\u003eet al.\u003c/em\u003e An intraoperative artificial intelligence system identifying anatomical landmarks for laparoscopic cholecystectomy: a prospective clinical feasibility trial (J-SUMMIT-C-01). Surg Endosc 37, 1933\u0026ndash;1942 (2023). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00464-022-09678-w\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00464-022-09678-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTokuyasu, T. \u003cem\u003eet al.\u003c/em\u003e Development of an artificial intelligence system using deep learning to indicate anatomical landmarks during laparoscopic cholecystectomy. Surg Endosc 35, 1651\u0026ndash;1658 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1007/s00464-020-07548-x\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1007/s00464-020-07548-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRo\u0026szlig;, T. \u003cem\u003eet al.\u003c/em\u003e Comparative validation of multi-instance instrument segmentation in endoscopy: Results of the ROBUST-MIS 2019 challenge. Med Image Anal 70, 101920 (2021). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1016/j.media.2020.101920\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1016/j.media.2020.101920\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLoukas, C., Gazis, A. \u0026amp; Schizas, D. Multiple instance convolutional neural network for gallbladder assessment from laparoscopic images. Int J Med Robot 18, e2445 (2022). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1002/rcs.2445\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1002/rcs.2445\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeibetseder, A., Schoeffmann, K., Keckstein, J. \u0026amp; Keckstein, S. \u003cem\u003ePost-surgical Endometriosis Segmentation in Laparoscopic Videos\u003c/em\u003e. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSonsilphong, S., Sonsilphong, A., Hormdee, D. \u0026amp; Khampitak, K. in \u003cem\u003e2022 International Electrical Engineering Congress (iEECON).\u003c/em\u003e 1\u0026ndash;4.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaballas, K., Bolingot, H. J., Libatique, N. \u0026amp; Tangonan, G. \u003cem\u003eDevelopment of a Visual Guidance System for Laparoscopic Surgical Palpation using Computer Vision\u003c/em\u003e. (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShorten, C. \u0026amp; Khoshgoftaar, T. M. A survey on Image Data Augmentation for Deep Learning. Journal of Big Data 6, 60 (2019). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org:10.1186/s40537-019-0197-0\u003c/span\u003e\u003cspan address=\"https://doi.org:10.1186/s40537-019-0197-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-5472618/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5472618/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eLaparoscopic pancreatic surgery remains highly challenging due to the complexity of the pancreas and surrounding vascular structures, with risk of injuring critical blood vessels such as the Superior Mesenteric Vein (SMV)-Portal Vein (PV) axis and splenic vein. Here, we evaluated the High Resolution Network (HRNet)-Full Convolutional Network (FCN) model for its ability to accurately identify vascular contours and improve surgical safety. Using 12,694 images from 126 laparoscopic distal pancreatectomy (LDP) videos and 35,986 images from 138 Whipple procedure videos, the model demonstrated robust performance, achieving a mean Dice coefficient of 0.754, a recall of 85.00%, and a precision of 91.10%. By combining datasets from LDP and Whipple procedures, the model showed strong generalization across different surgical contexts and achieved real-time processing speeds of 11 frames per second. These findings highlight the potential of HRNet-FCN to recognize anatomical landmarks, enhance surgical precision, reduce complications, and improve outcomes in laparoscopic pancreatic procedures.\u003c/p\u003e","manuscriptTitle":"Deep Learning HRNet-FCN for Blood Vessel Identification in Laparoscopic Pancreatic Surgery","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-09 17:22:59","doi":"10.21203/rs.3.rs-5472618/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-03-12T10:36:46+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-02-18T18:33:55+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"218229132570067087682117940644422614873","date":"2025-02-15T18:11:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-12-12T15:59:22+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"115172762403888796072322286377280161150","date":"2024-12-07T15:26:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"79981104855420355657074943042498451038","date":"2024-12-03T06:00:37+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"75531954927022774791063478373028762524","date":"2024-11-27T10:58:56+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-25T04:49:28+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-20T01:21:46+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-19T05:08:34+00:00","index":"","fulltext":""},{"type":"submitted","content":"npj Digital Medicine","date":"2024-11-18T04:21:10+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"npj-digital-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"npjdigitalmed","sideBox":"Learn more about [npj Digital Medicine](http://www.nature.com/npjdigitalmed/)","snPcode":"41746","submissionUrl":"https://submission.springernature.com/new-submission/41746/3","title":"npj Digital Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"NPJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"40035e89-1082-41d6-8cf8-927cf7b7ba4c","owner":[],"postedDate":"December 9th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":41300215,"name":"Biological sciences/Computational biology and bioinformatics/Image processing"},{"id":41300216,"name":"Biological sciences/Computational biology and bioinformatics/Machine learning"},{"id":41300217,"name":"Health sciences/Diseases/Gastrointestinal diseases/Pancreatic disease"},{"id":41300218,"name":"Health sciences/Diseases/Gastrointestinal diseases/Pancreatic disease/Pancreatic cancer"},{"id":41300219,"name":"Health sciences/Health care/Therapeutics/Surgery"},{"id":41300220,"name":"Health sciences/Health care/Therapeutics/Surgery/Surgical oncology"}],"tags":[],"updatedAt":"2025-05-05T16:04:47+00:00","versionOfRecord":{"articleIdentity":"rs-5472618","link":"https://doi.org/10.1038/s41746-025-01663-6","journal":{"identity":"npj-digital-medicine","isVorOnly":false,"title":"npj Digital Medicine"},"publishedOn":"2025-05-01 15:56:58","publishedOnDateReadable":"May 1st, 2025"},"versionCreatedAt":"2024-12-09 17:22:59","video":"","vorDoi":"10.1038/s41746-025-01663-6","vorDoiUrl":"https://doi.org/10.1038/s41746-025-01663-6","workflowStages":[]},"version":"v1","identity":"rs-5472618","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5472618","identity":"rs-5472618","version":["v1"]},"buildId":"zQwnuV7TCBrMSSSToR1PI","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}