Anomaly detection in fundus images by self-adaptive decomposition via local and color based sparse coding

Area of Science:

• Ophthalmology
• Medical Imaging
• Computer Vision

Background:

• Anomaly detection in color fundus images is difficult due to varied anomalies.
• Existing methods focus on background images, neglecting anomaly-specific features.

Purpose of the Study:

• To propose a novel multi-perspective anomaly modeling strategy for color fundus images.
• To enhance anomaly detection by simultaneously utilizing sequential sparsity and local/color saliency of anomalies.

Main Methods:

• Developed a simultaneous modeling strategy incorporating sequential sparsity and local/color saliency.
• Employed a Schatten p-norm metric for improved learning of heterogeneous background images.
• Enabled better discernment of anomalies from complex backgrounds.

Main Results:

• The proposed method demonstrates superior performance compared to existing techniques.
• Experimental comparisons validate the effectiveness of the multi-perspective approach.
• Improved accuracy in identifying diverse anomalies within fundus images.

Conclusions:

• The simultaneous modeling strategy effectively captures anomaly characteristics and background complexities.
• The Schatten p-norm metric enhances the ability to distinguish anomalies.
• This method offers a more robust and accurate solution for anomaly detection in retinal imaging.