Algorithm 1 PCA-based coarse noise reduction for SS-OCT fine-tuning.
From: Self-supervised model-informed deep learning for low-SNR SS-OCT domain transformation
Input: B-scans of individual images from SS-OCT dataset Output: Coarse noise-reduced, self-supervised noise-less images 1. Input Transformation:         • Align B-scans of each individual along a single axis.         • Reshape the resulting 3D volume into a 2D matrix of size (45000/60000 × N), where each row is a pixel vector and N is the number of B-scans. 2. Apply PCA:         • Perform PCA on the 2D matrix to identify the principal components.         • Sort components by descending energy (variance explained). 3. Component Selection:         • Retain only the top 15% of principal components with the highest energy. 4. Reconstruction:         • Reconstruct the B-scans using the selected principal components.         • Generate a noise-reduced image by projecting the original matrix onto the reduced component space. 5. Output:         • Use the reconstructed, coarse noise-reduced image as a “noise-less” target for self-supervised fine-tuning. |
