Summary of Content-based Image Retrieval For Multi-class Volumetric Radiology Images: a Benchmark Study, by Farnaz Khun Jush et al.
Content-Based Image Retrieval for Multi-Class Volumetric Radiology Images: A Benchmark Study
by Farnaz Khun Jush, Steffen Vogler, Tuan Truong, Matthias Lenga
First submitted to arxiv on: 15 May 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Information Retrieval (cs.IR)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This study tackles the challenge of content-based image retrieval (CBIR) in medical images, which is hindered by the 3D nature of these images. The authors draw upon recent research using pre-trained vision embeddings for radiology image retrieval and establish a benchmark for retrieving 3D volumetric medical images. A new dataset, TotalSegmentator (TS), with detailed multi-organ annotations is introduced, providing region-based and localized multi-organ retrieval capabilities. The study compares embeddings derived from pre-trained supervised models on medical images to those from unsupervised models on non-medical images for various anatomical structures at volume and region levels. The proposed late interaction re-ranking method achieves a high retrieval recall of 1.0 for diverse anatomical regions with a wide size range. This paper provides valuable insights and benchmarks for the development and evaluation of CBIR approaches in medical imaging, including region-based and localized multi-organ retrieval. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The researchers are trying to make it easier to find specific parts of medical images using computer algorithms. Right now, this task is hard because medical images have many dimensions (like layers). The team took inspiration from previous work that used pre-trained models for radiology image retrieval and created a new dataset with detailed labels for different organs. They tested how well different types of models performed on this dataset, comparing ones trained specifically for medical images to those trained on non-medical images. The results show that their new method works really well, allowing doctors to find the right parts of medical images quickly and accurately. |
Keywords
» Artificial intelligence » Recall » Supervised » Unsupervised
