Dataset: https://odir2019.grand-challenge.org/Download/

Abstract: Of late, usage of neural network in the field of disease detection has been on advanced stage. Hence, ocular disease diagnosis has also been under the influence of machine learning. Human eye is very prone to disorders like cataract, glaucoma, myopia etc. and with the passage of time, these diseases get more and more complex and the vision of human eye gradually decays. As a matter of fact, their early detection is mandatory in order to preclude complete blindness. Several diagnosis tests like visual acuity test, retinal exam, ocular tonometry are undertaken in real life but these are undoubtedly time consuming and frustrating for the patient as well. In this paper, a unique method for detecting eight types of ocular diseases using convolutional neural network (CNN) has been presented and its performance is evaluated. The affected regions for some diseases can also be detected. Some conventional pre-processing are performed and the data is sent to the network for rigorous classification. The model has been trained and tested with high-end graphics processing unit (GPU) on a brand-new dataset. Our developed model has achieved a cogent F-score of approx. 85%, Kappa score of 31% and an AUC value of 80.5%. Since this is the first “real-life” (i.e. plausible for a clinical scenario of patients including camera variation) prediction of multiple diseases in an eye based on this dataset, there is hardly any analogous task to compare with. So, our model has also been performed on other dataset and it has excelled with convincing F-score, Kappa score and an AUC value.

@INPROCEEDINGS{9065162,
  author={Islam, Md. Tariqul and Imran, Sheikh Asif and Arefeen, Asiful and Hasan, Mahmudul and Shahnaz, Celia},
  booktitle={2019 IEEE International Conference on Signal Processing, Information, Communication & Systems (SPICSCON)}, 
  title={Source and Camera Independent Ophthalmic Disease Recognition from Fundus Image Using Neural Network}, 
  year={2019},
  volume={},
  number={},
  pages={59-63},
  doi={10.1109/SPICSCON48833.2019.9065162}}