Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest

The lack of haptic feedback in robotic surgical systems can lead to unintended tissue damage as a result of excessive mechanical force. To address this issue, many studies in vision-based force sensing have focused on supervised learning approaches, which require labeled force data for training. How...

Full description

Saved in:
Bibliographic Details
Main Authors: Wenhui Zhuang, Kimihiko Masui, Naoto Kume, Megumi Nakao
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11097314/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849771869695115264
author Wenhui Zhuang
Kimihiko Masui
Naoto Kume
Megumi Nakao
author_facet Wenhui Zhuang
Kimihiko Masui
Naoto Kume
Megumi Nakao
author_sort Wenhui Zhuang
collection DOAJ
description The lack of haptic feedback in robotic surgical systems can lead to unintended tissue damage as a result of excessive mechanical force. To address this issue, many studies in vision-based force sensing have focused on supervised learning approaches, which require labeled force data for training. However, obtaining force labels in real surgical scenarios remains challenging. In this study, we propose an alternative approach by formulating the force feedback problem as an image-based anomaly detection task. By leveraging unsupervised learning, we overcome the limitations posed by the scarcity of labeled abnormal force data and present a novel abnormal force anomaly detection method. Our method employs a bidirectional generative adversarial network (BiGAN) and integrates a region-of-interest (ROI) guided module to enhance attention on the contact area between surgical instruments and tissue. The model is trained to distinguish between normal and abnormal force samples based on deformation features in an organ. Extensive experiments on an ex vivo porcine spleen manipulation dataset demonstrate the efficacy of the proposed ROI-guided BiGAN in accurately detecting abnormal forces. The ROI-guided BiGAN exhibits an area under the curve of 0.7963 and a recall rate of 0.9174 when applied to the test data. The proposed framework is expected to be useful as a practical tool for abnormal forceps force detection and to serve as a useful alarm in robot-assisted surgery.
format Article
id doaj-art-1c5b04b47bfa4c21ab1b8fb2909d3e05
institution DOAJ
issn 2169-3536
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Access
spelling doaj-art-1c5b04b47bfa4c21ab1b8fb2909d3e052025-08-20T03:02:29ZengIEEEIEEE Access2169-35362025-01-011313397113398310.1109/ACCESS.2025.359296011097314Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of InterestWenhui Zhuang0https://orcid.org/0009-0001-6444-6576Kimihiko Masui1Naoto Kume2https://orcid.org/0000-0001-8262-4754Megumi Nakao3https://orcid.org/0000-0002-5508-4366Graduate School of Informatics, Kyoto University, Kyoto, JapanGraduate School of Medicine, Kyoto University, Kyoto, JapanGraduate School of Medicine, Kyoto University, Kyoto, JapanGraduate School of Medicine, Kyoto University, Kyoto, JapanThe lack of haptic feedback in robotic surgical systems can lead to unintended tissue damage as a result of excessive mechanical force. To address this issue, many studies in vision-based force sensing have focused on supervised learning approaches, which require labeled force data for training. However, obtaining force labels in real surgical scenarios remains challenging. In this study, we propose an alternative approach by formulating the force feedback problem as an image-based anomaly detection task. By leveraging unsupervised learning, we overcome the limitations posed by the scarcity of labeled abnormal force data and present a novel abnormal force anomaly detection method. Our method employs a bidirectional generative adversarial network (BiGAN) and integrates a region-of-interest (ROI) guided module to enhance attention on the contact area between surgical instruments and tissue. The model is trained to distinguish between normal and abnormal force samples based on deformation features in an organ. Extensive experiments on an ex vivo porcine spleen manipulation dataset demonstrate the efficacy of the proposed ROI-guided BiGAN in accurately detecting abnormal forces. The ROI-guided BiGAN exhibits an area under the curve of 0.7963 and a recall rate of 0.9174 when applied to the test data. The proposed framework is expected to be useful as a practical tool for abnormal forceps force detection and to serve as a useful alarm in robot-assisted surgery.https://ieeexplore.ieee.org/document/11097314/Anomaly detectiondeep learningrobot-assisted surgery systemvision-based force sensing
spellingShingle Wenhui Zhuang
Kimihiko Masui
Naoto Kume
Megumi Nakao
Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
IEEE Access
Anomaly detection
deep learning
robot-assisted surgery system
vision-based force sensing
title Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
title_full Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
title_fullStr Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
title_full_unstemmed Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
title_short Unsupervised Anomaly Detection of Forceps Force by Localizing the Region of Interest
title_sort unsupervised anomaly detection of forceps force by localizing the region of interest
topic Anomaly detection
deep learning
robot-assisted surgery system
vision-based force sensing
url https://ieeexplore.ieee.org/document/11097314/
work_keys_str_mv AT wenhuizhuang unsupervisedanomalydetectionofforcepsforcebylocalizingtheregionofinterest
AT kimihikomasui unsupervisedanomalydetectionofforcepsforcebylocalizingtheregionofinterest
AT naotokume unsupervisedanomalydetectionofforcepsforcebylocalizingtheregionofinterest
AT meguminakao unsupervisedanomalydetectionofforcepsforcebylocalizingtheregionofinterest