Low-loss Low-rank Representation for Medical Image Analysis
Nan Zhang, Maolong Xi · Recent Patents on Engineering · 2025
Background: There has been extensive research on low-rank representation methods, and we firmly believe that this approach holds tremendous potential in addressing image-related challenges. Objective: We aim to preserve both the structure and low-rank information hidden in images of medical image databases to realize low-loss low-rank representation (LLRR). Method: We present low-loss low-rank representation(LLRR) to build a low-loss low-rank classifier (LLC) and a low-loss Low-rank discriminant projection(LLDP). Results: The LLC algorithm exhibits a disease classification accuracy that is superior, with a maximum difference of 10%, compared to other algorithms. After pairing the feature extractors with a Nearest Neighbor Classifier, LLDP achieved a maximum 5.6% advantage in disease classification accuracy. In other words, feature extraction and classification methods can be constructed based on the representation method proposed, and the performance of LLRR is superior. Conclusion: The experimental results indicate that LLDP and LLC consistently outperform other state-of-the-art methods, thus establishing LLRR as an effective data representation method