Kazuma Sato1,2, Takeo Fujita3, Hiroki Matsuzaki4, Nobuyoshi Takeshita4, Hisashi Fujiwara1, Shuichi Mitsunaga2,5,6, Takashi Kojima7, Kensaku Mori8, Hiroyuki Daiko9. 1. Division of Esophageal Surgery, National Cancer Center Hospital East, 6-5-1 Kashiwanoha, Kashiwa, Chiba, 277-8577, Japan. 2. Course of Advanced Clinical Research of Cancer, Juntendo University Graduate School of Medicine, Tokyo, 163-8001, Japan. 3. Division of Esophageal Surgery, National Cancer Center Hospital East, 6-5-1 Kashiwanoha, Kashiwa, Chiba, 277-8577, Japan. takfujit@east.ncc.go.jp. 4. Surgical Device Innovation Office, National Cancer Center Hospital East, Kashiwa, Chiba, Japan. 5. Division of Hepatobiliary and Pancreatic Oncology, National Cancer Center Hospital East, Kashiwa, Chiba, Japan. 6. Division of Biomarker Discovery, Exploratory Oncology Research and Clinical Trial Center, National Cancer Center, Kashiwa, Chiba, Japan. 7. Division of Gastrointestinal Oncology, National Cancer Center Hospital East, Kashiwa, Japan. 8. Graduate School of Informatics, Nagoya University School of Medicine, Nagoya, Aichi, Japan. 9. Division of Esophageal Surgery, National Cancer Center Hospital, Tokyo, Japan.
Abstract
BACKGROUND: Artificial intelligence (AI) has been largely investigated in the field of surgery, particularly in quality assurance. However, AI-guided navigation during surgery has not yet been put into practice because a sufficient level of performance has not been reached. We aimed to develop deep learning-based AI image processing software to identify the location of the recurrent laryngeal nerve during thoracoscopic esophagectomy and determine whether the incidence of recurrent laryngeal nerve paralysis is reduced using this software. METHODS: More than 3000 images extracted from 20 thoracoscopic esophagectomy videos and 40 images extracted from 8 thoracoscopic esophagectomy videos were annotated for identification of the recurrent laryngeal nerve. The Dice coefficient was used to assess the detection performance of the model and that of surgeons (specialized esophageal surgeons and certified general gastrointestinal surgeons). The performance was compared using a test set. RESULTS: The average Dice coefficient of the AI model was 0.58. This was not significantly different from the Dice coefficient of the group of specialized esophageal surgeons (P = 0.26); however, it was significantly higher than that of the group of certified general gastrointestinal surgeons (P = 0.019). CONCLUSIONS: Our software's performance in identification of the recurrent laryngeal nerve was superior to that of general surgeons and almost reached that of specialized surgeons. Our software provides real-time identification and will be useful for thoracoscopic esophagectomy after further developments.
BACKGROUND: Artificial intelligence (AI) has been largely investigated in the field of surgery, particularly in quality assurance. However, AI-guided navigation during surgery has not yet been put into practice because a sufficient level of performance has not been reached. We aimed to develop deep learning-based AI image processing software to identify the location of the recurrent laryngeal nerve during thoracoscopic esophagectomy and determine whether the incidence of recurrent laryngeal nerve paralysis is reduced using this software. METHODS: More than 3000 images extracted from 20 thoracoscopic esophagectomy videos and 40 images extracted from 8 thoracoscopic esophagectomy videos were annotated for identification of the recurrent laryngeal nerve. The Dice coefficient was used to assess the detection performance of the model and that of surgeons (specialized esophageal surgeons and certified general gastrointestinal surgeons). The performance was compared using a test set. RESULTS: The average Dice coefficient of the AI model was 0.58. This was not significantly different from the Dice coefficient of the group of specialized esophageal surgeons (P = 0.26); however, it was significantly higher than that of the group of certified general gastrointestinal surgeons (P = 0.019). CONCLUSIONS: Our software's performance in identification of the recurrent laryngeal nerve was superior to that of general surgeons and almost reached that of specialized surgeons. Our software provides real-time identification and will be useful for thoracoscopic esophagectomy after further developments.
Authors: Pietro Mascagni; Armine Vardazaryan; Deepak Alapatt; Takeshi Urade; Taha Emre; Claudio Fiorillo; Patrick Pessaux; Didier Mutter; Jacques Marescaux; Guido Costamagna; Bernard Dallemagne; Nicolas Padoy Journal: Ann Surg Date: 2020-11-16 Impact factor: 13.787
Authors: Amin Madani; Babak Namazi; Maria S Altieri; Daniel A Hashimoto; Angela Maria Rivera; Philip H Pucher; Allison Navarrete-Welton; Ganesh Sankaranarayanan; L Michael Brunt; Allan Okrainec; Adnan Alseidi Journal: Ann Surg Date: 2020-11-13 Impact factor: 13.787