https://doi.org/10.1140/epjs/s11734-024-01338-5
Regular Article
A new data label conversion algorithm for YOLO segmentation of medical images
1
Institute of Natural Sciences, Sakarya University, Sakarya, Türkiye
2
Computer Engineering, Sakarya University of Applied Sciences, Sakarya, Türkiye
3
Electrical and Electronics Engineering, Sakarya University of Applied Sciences, Sakarya, Türkiye
4
Software Engineering, Sakarya University, Sakarya, Türkiye
5
Computer Engineering, Gaziantep Islam Science And Technology University, Gaziantep, Türkiye
a
muhammedtelceken@subu.edu.tr
Received:
31
May
2024
Accepted:
5
September
2024
Published online:
2
October
2024
Object detection and segmentation architectures like YOLO have increased accuracy and speed, making them more practical in medical applications, including polyp identification for medical imaging and diagnosis. However, accurately defining segmentation boundaries to train a YOLO model for medical diagnosis requires considerable manual effort. This study developed a new method for automatically generating labels from ground-truth masks to process datasets for training YOLO-based segmentation algorithms. The developed method offers a simple interface for users to readily sample segmentation masks in datasets into labels compatible with the YOLO format. The technique involves performing contour extraction while processing labeled masks to ensure that the resulting contour information represents the object boundaries and makes it possible to map them according to the YOLO format. The efficiency of the method has been shown via training YOLOv5, YOLOv7, and YOLOv8-based models. Comparative results on the polyp segmentation dataset show successful improvements using YOLO-based segmentation over the literature.
Copyright comment Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
© The Author(s), under exclusive licence to EDP Sciences, Springer-Verlag GmbH Germany, part of Springer Nature 2024. Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
