Detecting Medical Equipment in the Catheterization Laboratory using Computer Vision

Detecting Medical Equipment in the Catheterization Laboratory using Computer Vision

Dai, Renjie (TU Delft Electrical Engineering, Mathematics and Computer Science)

Dauwels, J.H.G. (mentor)
van Gemert, J.C. (graduation committee)

Delft University of Technology

Electrical Engineering | Circuits and Systems

2022-09-30

Workflow analysis aims to improve the efficiency and safety in operating rooms by analysing surgical processes and providing feedback or support, where observations can be made and evaluated by algorithms rather than human experts. For our study, we mount five calibrated cameras from different angles in a Catheterization Laboratory (Cath Lab) to observe and analyse Cardiac Angiogram procedures.

To automate the classification of workflow and personnel activities, we propose an object detection algorithm based on Scaled-YOLOv4 with a filter to improve bounding box prediction. Scaled-YOLOv4, as a state-of-the-art technique, is featured with extremely fast processing speed and decent precision. We improve the Scaled-YOLOv4 network by using different IoU losses and an additional transformer layer.

In addition, we find that Scaled-YOLOv4 still suffers the object occlusion problem, especially in Cath Lab with limited room space but massive medical equipment. This can result in the inaccurate prediction of bounding boxes. In this work, we also develop a filter following Scaled-YOLOv4 to improve the prediction of bounding box by matching the features detected from different cameras. With the keypoints detected by SuperPoint and matched by SuperGLue, the filter adjusts the boundaries of bounding box to include all the matched keypoints and exclude unmatched points.

The proposed algorithm achieves 95.1$\%$ mAP in detecting medical equipment in Catheterization Laboratory and a real-time speed, 58 FPS on RTX 3090.

Object detection
Catheterization laboratory
Multiple Cameras
Medical equipment
feature matching

http://resolver.tudelft.nl/uuid:1894799d-302b-44bc-841e-bf652471330b

Student theses

master thesis

© 2022 Renjie Dai