Incorporating weak anomaly labels into standard segmentation models improves lesion segmentation results without requiring extra manual labels, enhancing the potential of anomaly guided segmentation for retinal optical coherence tomography scans.
The paper evaluates domain generalization strategies for optic disc and cup segmentation in fundus images, highlighting issues with existing methods when applied to uncropped images, and proposes a semi-supervised learning approach based on the Noisy Student framework to improve performance across diverse datasets.
This paper introduces PALM, an open fundus imaging dataset for pathological myopia recognition, featuring 1200 images with associated labels for pathologic myopia category and manual annotations of optic disc, fovea position, and lesions like patchy retinal atrophy and retinal detachment, aiding in automated diagnostic tools development.
The GAMMA Challenge addresses the need for multi-modality glaucoma grading, providing a dataset with both 2D fundus images and 3D OCT volumes, inviting algorithm development and evaluation, leading to practical insights for clinical diagnosis.
With no other methodological innovation than a carefully designed training, our ResNet model achieved an AUC = 0.955 (0.953 - 0.956) on a combined test set of 61007 test images from different public datasets, which is in line or even better than what other more complex deep learning models reported in the literature.
We experimentally validate whether using coarse-to-fine models instead of one-stage models is appropriate or not for segmenting the optic disc and the optic cup in color fundus images. We observed that models trained with the right amount of data can perform much better than coarse-to-find approaches.
We introduced a fully automated approach to segment the photorceptor layer, evaluate its thickness and track potential disruptions using an ensemble of deep neural networks.