- Title
- 3D-CNN for glaucoma detection using optical coherence tomography
- Creator
- George, Yasmeen; Antony, Bhavna; Ishikawa, Hiroshi; Wollstein, Gadi; Schuman, Joel; Garnavi, Rahil
- Date
- 2019
- Type
- Text; Conference proceedings
- Identifier
- http://researchonline.federation.edu.au/vital/access/HandleResolver/1959.17/197418
- Identifier
- vital:18840
- Identifier
- ISBN:978-3-030-32956-3
- Abstract
- The large size of raw 3D optical coherence tomography (OCT) volumes poses challenges for deep learning methods as it cannot be accommodated on a single GPU in its original resolution. The direct analysis of these volumes however, provides advantages such as circumventing the need for the segmentation of retinal structures. Previously, a deep learning (DL) approach was proposed for the detection of glaucoma directly from 3D OCT volumes, where the volumes were significantly downsampled first. In this paper, we propose an end-to-end DL model for the detection of glaucoma that doubles the number of input voxels of the previously proposed method, and also boasts an improved AUC = 0.973 over the results obtained using the previously proposed approach of AUC = 0.946. Furthermore, this paper also includes a quantitative analysis of the regions of the volume highlighted by grad-CAM visualization. Occlusion of these highlighted regions resulted in a drop in performance by 40%, indicating that the regions highlighted by gradient-weighted class activation maps (grad-CAM) are indeed crucial to the performance of the model.
- Publisher
- Springer International Publishing
- Relation
- Ophthalmic Medical Image Analysis 6th International workshop, OMIA; Shenzen, China; October 17, 2019 in Lecture Notes in Computer Science (LNCS, volume 11855) p. 52-59
- Rights
- All metadata describing materials held in, or linked to, the repository is freely available under a CC0 licence
- Rights
- Copyright Springer
- Reviewed
- Hits: 103
- Visitors: 94
- Downloads: 0