CSCR OCT Analyzer & Longitudinal Clinical Study Pipeline

A MATLAB-based image processing, machine learning, and longitudinal clinical analytics pipeline designed to analyze Optical Coherence Tomography (OCT) retinal scans. The system integrates traditional computer vision techniques with the Segment Anything Model (SAM) to automatically align scans, segment retinal layers, detect macular holes, extract speckle statistics from Regions of Interest (ROIs), and track disease evolution across clinical visits through a persistent local database and an interactive dashboard.

---

Key Features

• Automated Pre-processing & Tilt Correction: Resizes (992x1536), estimates scan tilt via rough ILM detection, flattens the scan, crops to a central 6mm zone, and extracts baseline SAM embeddings.
• SAM-Powered Layer Segmentation:
  • ILM Segmentation: Prompts SAM to segment and trace the Inner Limiting Membrane.
  • Macular Hole Detection: Vertically stretches the image, applies morphological thresholding and region properties to identify holes, segments with SAM, and inversely maps back to the aligned space.
  • RPE & Inner Retinal Tracks: Segments top and bottom boundaries of the Retinal Pigment Epithelium (RPE) into 3 sub-layers and divides the inner retina into 4 distinct tracks using local intensity peak tracking.
• Dynamic ROI Speckle Analysis:
  • Hole present: Dynamically generates 5 Left ROIs and 5 Right ROIs with safety margins.
  • No hole: Generates 5 Full-width continuous ROIs.
  • Extracts Contrast Ratio ($\text{CR} = \sigma / \mu$), Densitometry, and distribution fits (Burr-2, Gamma, K-distribution, Weibull, etc.) via speckle_stats.m.
• Zero-Setup Local Database (db_manager.m):
  • Fully native, persistent table database (oct_study_db.mat) with no SQL servers or extra toolboxes required.
  • Maintains relational tables for Visits, Scans, and VisitStats.
• Longitudinal Statistical Analytics Engine (compute_statistics.m):
  • Computes the $L4/L5$ speckle contrast ratio.
  • Computes exact daily rate of change ($\Delta\text{ratio}/\Delta\text{day}$) between consecutive clinical visits.
  • Computes visit aggregates (means and standard deviations across slices).
  • Evaluates Pearson and Spearman clinical correlations between Central Retinal Thickness ($\mu m$) and speckle contrast ratios.
  • Automatically renders publication-grade two-panel figures with clinical status bands (Active Flare-up, Pre-recurrence, Remission).
• Interactive MATLAB UI Dashboard (oct_dashboard.m):
  • Desktop application built with MATLAB uifigure.
  • Visualizes temporal time-series, dual-axis thickness overlays, and rate-of-change velocity bars.
  • Interactive visit selector, central thickness editor, clinical status badges, and visit creation (e.g. Visit 23+).
• Flexible Single & Batch Execution:
  • Process single folders or batch-process entire directory trees.
  • Features skip/resume capability and reuses the instantiated SAM MODEL in memory to maximize processing speed.
• One-Click Excel Export & Migration:
  • Importer to migrate historical visits from Statistics.xlsx.
  • Multi-sheet Excel exporter for sharing audit data with clinical collaborators and thesis supervisors.

---

Project Structure

├── main_OCT_retinal_scan_script.m   # Entry-point runner for single folder processing and dashboard launch
├── oct_dashboard.m                  # Interactive MATLAB UI Desktop Dashboard (uifigure)
├── db_manager.m                     # Persistent local database manager (oct_study_db.mat)
├── compute_statistics.m             # Analytics engine (L4/L5 ratios, rates of change, correlations, figures)
├── process_oct_folder.m             # Modular folder runner with metadata parsing and DB recording
├── batch_process_folders.m          # Batch runner across all subdirectories with skip/resume logic
├── migrate_excel_to_db.m            # Historical data importer from Statistics.xlsx into the database
├── preprocess_OCT.m                 # Standardizes scan size, flattens tilt, crops 6mm, extracts SAM embeddings
├── segment_ILM.m                    # Segments the Inner Limiting Membrane (ILM) using SAM prompts
├── detect_hole.m                    # Stretches scan vertically, detects macular holes, and segments boundary
├── segment_layers.m                 # Maps RPE top/bottom and 4 inner retinal blue tracks
├── analyze_rois.m                   # Slices image into 15 ROIs (Left, Right, Full) and calls speckle_stats
├── speckle_stats.m                  # Calculates contrast ratios, densitometry, and 11 distribution fits
├── align_image.m                    # Flattens tilted scans with synthetic noise padding
├── burrfit2.m / burr2pdf.m          # Burr-2 distribution parameter estimation and PDF evaluation
├── kfit.m / kpdf.m                  # K-distribution parameter estimation and PDF evaluation
├── fake_data.m                      # Synthetic benchmark script for testing clinical status bands
├── GEMINI.md                        # Architectural context, data flow, and development guidelines
└── .gitignore                       # Medical data protection rules (blocks scans, DBs, and spreadsheets)

---

Prerequisites

• MATLAB (Tested on R2025b)
• Required Toolboxes:
  • Image Processing Toolbox
  • Computer Vision Toolbox
  • Statistics and Machine Learning Toolbox
  • Deep Learning Toolbox
  • MATLAB Support Package for Segment Anything Model (SAM)

(Note: The Database Toolbox is NOT required; db_manager.m uses high-performance native MATLAB table persistence).

---

Getting Started

1. Launch the Interactive Dashboard

To inspect existing visits, view longitudinal graphs, edit central retinal thicknesses, or export Excel reports:

oct_dashboard

2. Process a Single OCT Scan Folder

To analyze an individual visit folder using the graphical file selector:

main_OCT_retinal_scan_script

Or run it programmatically:

folderPath = 'C:\Path\To\Scan_Folder';
process_oct_folder(folderPath);

3. Batch Process Multiple Folders

To scan a root directory and process all unrecorded visit folders automatically:

batch_process_folders('C:\Path\To\Parent_Data_Folder');

Folders already recorded in the database are automatically skipped to save time.

4. Generate Clinical Summary & Correlation Reports

To print the full statistical table and correlation coefficients in the MATLAB Command Window:

compute_statistics('report');

To generate and save the publication-grade multi-panel figure:

compute_statistics('plot', 'longitudinal_study.png');

5. Export Data to Excel

To generate a consolidated multi-sheet Excel file for your supervisor:

db_manager('export_excel', 'OCT_Study_Export.xlsx');

---

Medical Data Privacy

To comply with patient data privacy regulations, no raw scans (.bmp, .png, .tif), local database files (oct_study_db.mat), or patient spreadsheets (*.xlsx) are committed to version control. The repository includes a .gitignore file that strictly shields clinical data while tracking all code, functions, and documentation.