Anatomy Segmentation in Laparoscopic Surgery: Comparison of Machine Learning and Human Expertise
Anatomy Segmentation in Laparoscopic Surgery: Comparison of Machine Learning and Human Expertise
Structured Abstract BackgroundLack of anatomy recognition represents a clinically relevant risk in abdominal surgery. Machine learning (ML) methods can help identify visible patterns and risk structures, however, their practical value remains largely unclear. Materials and MethodsBased on a novel dataset of 13195 laparoscopic images with pixel-wise segmentations of eleven anatomical structures, we developed specialized segmentation models for each structure and a combined model for all anatomical structures, and compared segmentation performance of both algorithms to a cohort of 28 physicians, medical students, and medical laypersons using the example of pancreas segmentation. ResultsMean Intersection-over-Union for segmentation of intraabdominal structures ranged from 0.28 to 0.83 and from 0.23 to 0.77 for the structure-specific and the combined semantic segmentation model, respectively. At average inference times per model of 28 ms and 71 ms, respectively, both variants are capable of near-real-time operation. Both models outperformed 26 out of 28 human participants in pancreas segmentation. ConclusionsThese results demonstrate that ML methods have the potential to provide relevant assistance in anatomy recognition in minimally-invasive surgery in near-real-time. Future research should investigate the educational value and subsequent clinical impact of respective assistance systems. HighlightsBased on a novel large-scale dataset of 13195 laparoscopic images, two machine learning models were developed for automated identification and delineation of 11 anatomical structures: One model for each structure (abdominal wall, colon, intestinal vessels (inferior mesenteric artery and inferior mesenteric vein with their subsidiary vessels), liver, pancreas, small intestine, spleen, stomach, ureter and vesicular glands) and a combined model for all structures.The structure-specific and combined segmentation models demonstrated similar performance in identifying intraabdominal structures and can operate in near-real-time.Both models outperformed 26 out of 28 human participants in pancreas segmentation, demonstrating their potential for real-time assistance in recognizing anatomical landmarks during minimally-invasive surgery.
Distler Marius、Jenke Alexander C.、Speidel Stefanie、Leger Stefan、Kolbinger Fiona R.、Bodenstedt Sebastian、Carstens Matthias、Weitz J¨1rgen、Rinner Franziska M.
Department of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universit?t Dresden||National Center for Tumor Diseases (NCT/UCC) German Cancer Research Center (DKFZ) Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universit?t Dresden Helmholtz-Zentrum Dresden-Rossendorf (HZDR)Division of Translational Surgical Oncology, National Center for Tumor Diseases (NCT), Partner Site DresdenElse Kr?ner Fresenius Center for Digital Health (EKFZ), Technische Universit?t Dresden||Division of Translational Surgical Oncology, National Center for Tumor Diseases (NCT), Partner Site Dresden||Cluster of Excellence ?°Centre for Tactile Internet with Human-in-the-Loop?± (CeTI), Technische Universit?t DresdenElse Kr?ner Fresenius Center for Digital Health (EKFZ), Technische Universit?t Dresden||Division of Translational Surgical Oncology, National Center for Tumor Diseases (NCT), Partner Site DresdenDepartment of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universit?t Dresden||National Center for Tumor Diseases (NCT/UCC) German Cancer Research Center (DKFZ) Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universit?t Dresden Helmholtz-Zentrum Dresden-Rossendorf (HZDR)||Else Kr?ner Fresenius Center for Digital Health (EKFZ), Technische Universit?t DresdenDivision of Translational Surgical Oncology, National Center for Tumor Diseases (NCT), Partner Site Dresden||Cluster of Excellence ?°Centre for Tactile Internet with Human-in-the-Loop?± (CeTI), Technische Universit?t DresdenDepartment of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universit?t DresdenDepartment of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universit?t Dresden||National Center for Tumor Diseases (NCT/UCC) German Cancer Research Center (DKFZ) Faculty of Medicine and University Hospital Carl Gustav Carus, Technische Universit?t Dresden Helmholtz-Zentrum Dresden-Rossendorf (HZDR)||Else Kr?ner Fresenius Center for Digital Health (EKFZ), Technische Universit?t Dresden||Cluster of Excellence ?°Centre for Tactile Internet with Human-in-the-Loop?± (CeTI), Technische Universit?t DresdenDepartment of Visceral, Thoracic and Vascular Surgery, University Hospital and Faculty of Medicine Carl Gustav Carus, Technische Universit?t Dresden
医学研究方法临床医学外科学
Minimally-invasive surgerylaparoscopysurgical data sciencesurgical anatomysurgical innovationartificial intelligence
Distler Marius,Jenke Alexander C.,Speidel Stefanie,Leger Stefan,Kolbinger Fiona R.,Bodenstedt Sebastian,Carstens Matthias,Weitz J¨1rgen,Rinner Franziska M..Anatomy Segmentation in Laparoscopic Surgery: Comparison of Machine Learning and Human Expertise[EB/OL].(2025-03-28)[2025-05-28].https://www.medrxiv.org/content/10.1101/2022.11.11.22282215.点此复制
评论