Table 4.
Summarization of the pre-processing techniques for fundus images utilized in diagnosing retinal disease [31,69,70,72].
| Pre-Processing Technique | Description | Complexity | Effectiveness | Robustness | Ease of Implementation |
|---|---|---|---|---|---|
| Color Space Transformation | Enhances DL model performance by selectively utilizing a single color channel from the RGB Channels, often removing the green channel to eliminate visually rich information, particularly in high contrast images | Low | Moderate | High | High |
| Cumulative Distribution Function (CDF) | Simplifies understanding of image features and pixel intensity distribution through a cumulative probability distribution, aiding in identifying areas of interest | Low | High | Moderate | Moderate |
| Noise Removal | Eliminates unwanted noise using various denoising algorithms such as non-local means denoising, median filters, and Gaussian filters | Moderate | High | High | Moderate |
| Contrast Enhancement (CLAHE) | Widely used approach for enhancing contrast and addressing over-amplified contrast in certain pixel portions, improving the quality of fundus images for analysis | Moderate | High | High | High |
| Segmentation Mask | Utilizes binary masks to isolate regions of interest (ROIs) within fundus images, improving diagnostic accuracy by selecting specific regions for analysis while excluding background noise | Moderate | High | High | Moderate |
| Contour Analysis | Essential for fine-tuning ROIs and locating object boundaries in images, providing attributes like centroid, area, and perimeter for object modification and determination | Moderate | High | Moderate | Moderate |
| Augmentation | Balances image datasets through techniques like rotation, translation, flipping, and rescaling, enhancing model robustness and performance | Low | High | High | High |
| Cropping and Extracting ROI | Isolates significant areas within images for analysis, reducing unnecessary learning effort during model training | Low | Moderate | High | High |
| Histogram Equalization | Enhances overall contrast in fundus images, making background pixels stand out and improving image clarity. | Low | Moderate | Moderate | High |
| Resized Image | Maintains consistency across the dataset by resizing images to standard dimensions | Low | Low | High | High |
| Enhanced Image (Improving Contrast) | Lowers noise and enhances contrast in images, improving overall image quality | Moderate | High | Moderate | High |