What question did this study set out to answer?

The aim is to develop and validate a deep-learning model to detect and quantify intraoperative bleeding during unilateral biportal endoscopic spine surgery.

March 6, 2026Journal of Clinical Medicine1 citationsOpen Access

Deep-Learning Model for Automated Detection of Bleeding in Spine Surgery

Development of a Deep-Learning Model for Automated Detection and Quantification of Bleeding in Unilateral Biportal Endoscopic Spine Surgery

View Full Paper

Authors

TYTakaki YoshimizuSeirei Hamamatsu General Hospital DSDaisuke SakaiTokai University Hospital DMDaiki MoritaTokyo Denki University

Discussion

Member takes

Overview

Automated detection of intraoperative bleeding improves surgical outcomes in spine surgery, suggesting enhanced decision-making capability.

Key Points

The aim is to develop and validate a deep-learning model to detect and quantify intraoperative bleeding during unilateral biportal endoscopic spine surgery.
Extracted 223,568 still images from 20 UBE videos for model training.
Utilized a U-Net++ segmentation model based on HSV thresholding for red mask generation.
Fine-tuned the model using 350 manually annotated images to differentiate bleeding from non-bleeding regions.
Evaluated model performance against 180 ground-truth images annotated by spine surgeons.
Calculated Dice and intersection-over-union (IoU) scores, and performed correlation analyses on inter-annotator agreement.
The HSV-based model showed high fidelity in reproducing red regions but limited agreement with ground-truth bleeding (median Dice = 0.57, IoU = 0.40).
The fine-tuned model achieved an accuracy of 86% for binary classification of bleeding, with a sensitivity of 93% and specificity of 60%.
For pixel-level segmentation, the median Dice score improved to 0.79 and median IoU to 0.65 on ground-truth-positive images.

Ask AI

Helpful

Bookmark

View Full Paper

Cite This Study

Yoshimizu et al. (2026) studied this question.

synapsesocial.com/papers/69aa7096531e4c4a9ff5a805 https://doi.org/10.3390/jcm15051934

Ask AI

Helpful

Bookmark

View Full Paper

March 6, 2026Journal of Clinical Medicine1 citationsOpen Access

Deep-Learning Model for Automated Detection of Bleeding in Spine Surgery

Development of a Deep-Learning Model for Automated Detection and Quantification of Bleeding in Unilateral Biportal Endoscopic Spine Surgery

View Full Paper

Authors

TYTakaki YoshimizuSeirei Hamamatsu General Hospital DSDaisuke SakaiTokai University Hospital DMDaiki MoritaTokyo Denki University

Discussion

Member takes

Overview

Automated detection of intraoperative bleeding improves surgical outcomes in spine surgery, suggesting enhanced decision-making capability.

Key Points

The aim is to develop and validate a deep-learning model to detect and quantify intraoperative bleeding during unilateral biportal endoscopic spine surgery.
Extracted 223,568 still images from 20 UBE videos for model training.
Utilized a U-Net++ segmentation model based on HSV thresholding for red mask generation.
Fine-tuned the model using 350 manually annotated images to differentiate bleeding from non-bleeding regions.
Evaluated model performance against 180 ground-truth images annotated by spine surgeons.
Calculated Dice and intersection-over-union (IoU) scores, and performed correlation analyses on inter-annotator agreement.
The HSV-based model showed high fidelity in reproducing red regions but limited agreement with ground-truth bleeding (median Dice = 0.57, IoU = 0.40).
The fine-tuned model achieved an accuracy of 86% for binary classification of bleeding, with a sensitivity of 93% and specificity of 60%.
For pixel-level segmentation, the median Dice score improved to 0.79 and median IoU to 0.65 on ground-truth-positive images.

Ask AI

Helpful

Bookmark

View Full Paper

Cite This Study

Yoshimizu et al. (2026) studied this question.

synapsesocial.com/papers/69aa7096531e4c4a9ff5a805 https://doi.org/10.3390/jcm15051934

Ask AI

Helpful

Bookmark

View Full Paper