Efficient optic cup detection from intra-image learning with retinal structure priors

摘要: We present a superpixel based learning framework on retinal structure priors for glaucoma diagnosis. In digital fundus photographs, our method automatically localizes the optic cup, which is primary image component clinically used identifying glaucoma. This provides three major contributions. First, it proposes processing of images at level, leads to features more descriptive and effective than those employed by pixel-based techniques, while yielding significant computational savings over methods sliding windows. Second, classifier process does not rely pre-labeled training samples, but rather samples are extracted from test itself using structural relative cup disc positions. Third, we classification refinement scheme that utilizes both local context. Tested ORIGA−light clinical dataset comprised 650 images, proposed achieves 26.7% non-overlap ratio with manually-labeled ground-truth 0.081 absolute cup-to-disc (CDR) error, simple yet widely diagnostic measure. level accuracy comparable or higher state-of-the-art technique [1], speedup factor tens hundreds.