Title: ClassifyViStA: WCE Classification with Visual Understanding through Segmentation and Attention
Published Date: 2024-12-24
Abstract:Gastrointestinal bleeding is a serious medical condition that poses significant diagnostic challenges, especially in areas with limited medical resources. Wireless capsule endoscopy (WCE) has become a powerful diagnostic tool for visualizing the gastrointestinal tract, but it requires time-consuming manual analysis by experienced gastroenterologists, which is prone to human error and becomes inefficient with increasing patient numbers. To address this challenge, we propose ClassifyViStA, an AI-based framework designed to automatically detect and classify bleeding and non-bleeding frames from WCE videos. The model includes a standard classification path and adds two specialized branches: an implicit attention branch and a segmentation branch. The attention branch focuses on bleeding areas, while the segmentation branch generates accurate segmentation masks, which are used for classification and interpretability. The model is built on a combination of ResNet18 and VGG16 architectures to enhance classification performance. For the detection of bleeding areas, the soft non-maximum suppression (Soft NMS) method of YOLOv8 is implemented, improving the handling of overlapping bounding boxes for more accurate and nuanced detection. By using segmentation masks to interpret classification results, the system’s interpretability is enhanced, providing insights similar to how gastroenterologists identify bleeding areas, thus deepening the understanding of the decision-making process. This approach not only automates the detection of gastrointestinal bleeding but also provides an interpretable solution that alleviates the burden on medical professionals and improves diagnostic efficiency. The relevant code can be found at ClassifyViStA.
Paper link: http://arxiv.org/abs/2412.18591v1===========================================
Welcome to join myKnowledge Planet
Long press or scan the QR code below
Planet number mainly provides value-added services for public accountsincluding but not limited to
|
|
PhD team provides secondary discussions on public account papers,paper downloads,algorithm code guidance |
|
|
Face-to-face communication with authors of monographs,“Deep Learning from Theory to Practice”,“Rapid Deployment of Large Models LLM Strategies and Practices” authors face-to-face communication about book content. Follow the booksfor beginners from scratch. |
|
|
IT career guidance, large company interview guidance,pitfall guidance |
|
| Aggregation of large model/researcher information from major companies at home and abroad (Twitter, Facebook, YouTube) |
|
| RAG consulting, GraphRAG consulting, large model consulting |
|
|
Graph algorithm consulting, graph mining, graph recommendation Graph database consulting, neo4j, nebula |