Age-related macular degeneration (AMD) and diabetic retinopathy (DR) are leading causes of blindness worldwide, affecting millions. Early detection and treatment are crucial for these retinal diseases, and retinal optical coherence tomography (OCT) imaging is a key diagnostic tool. Automated image analysis using OCT images has the potential to improve diagnosis and treatment monitoring. However, challenges like noise, distortions, and intensity variations make accurate information extraction difficult. While successful algorithms exist, their robust use in clinical practice remains an ongoing research focus. Here, we present a new method for DR detection on OCT volumes using a common dataset and achieving high performance in retinal image analysis competition.

1. Train Data Preparation:

   • Run the Main_ROCC_Per_B_Scan.m script.
   • This script produces the FeatureVector_HOG.mat file.

2. Validation Data Preparation:

   • Run the Main_ROCC_Per_B_Scan_Valid.m script.
   • This script produces the FeatureVector_Valid_HOG.mat file.

3. Test Data Preparation:

   • Run the Main_ROCC_Per_B_Scan_Test.m script.
   • This script produces the FeatureVector_Test_HOG.mat file.

4. Classification:

   • Run one of the following scripts depending on your training and testing needs:
     • Main_ROCC_Per_B_Scan_Classification.m
     • Main_ROCC_Per_B_Scan_Classification_Valid.m
     • Main_ROCC_Per_B_Scan_Classification_Test.m