RetinaAI Platform v2.0 • Lesion Quantification & Groq AI Active

Transforming Retinal Diagnostic Intelligence with Clinical AI

Powered by checkpoints/best.pt GAN restoration, Retinal Lesion Quantification & Groq Vision AI inspection.

Explore Diagnostic Pillars
Model Checkpoint: checkpoints/best.pt
U-Net Generator • Lesion Quantification • Groq Vision AI
Model: best.pt Loaded Lesion Engine: Active
Restored Retina
Degraded Retina
↔
← Degraded Input Image best.pt Restored Output (with Vein Enhancement) →
best.pt
Model Checkpoint
Groq Vision
Llama 3.2 / 3.3
Lesion AI
Lesion Quantification
Epoch 34
GAN Checkpoint
Platform Architecture

Clinical Diagnostic Suite

An end-to-end diagnostic ecosystem powered by Lesion Quantification & Groq Vision AI.

🔬

1. GAN Neural Restoration

U-Net generator trained model (best.pt) with dedicated green-channel attention for noise removal and clarity restoration.

checkpoints/best.pt
📊

2. AI Change Analysis

Wavelet frequency decomposition & local SSIM maps that explain precisely what structures were recovered by AI.

Explainable AI
🔍

3. Groq Vision ROI Inspector

Click-to-inspect interactive ROI pin evaluation using Groq Vision AI with smooth Grad-CAM heatmap toggles.

Groq Vision AI
🎯

4. Retinal Lesion Quantification

Calculates total pathology lesion counts, affected retinal area percentage, and 4-quadrant lesion density breakdown.

Lesion Analytics
🧬

5. Retinal Biomarkers

Calculates vascular tortuosity, box-counting fractal dimension, AVR ratio, and image gradability index.

Biomarker Suite
🔥

6. Grad-CAM Attention

Gradient-weighted activation heatmaps displaying exact spatial focus during restoration.

Neural Heatmaps
👁️
RetinaAI Console
Model: checkpoints/best.pt
📸 Image Input
🔬 Restored Comparison
🔍 Groq Vision ROI Inspector
🧬 Biomarker Panel
🎯 Lesion Quantification
🧠 Diagnostic Report

Upload Retinal Image for Inference

Upload a fundus photograph to run checkpoints/best.pt GAN restoration and diagnostic analysis.

🧿
Click or drag retinal fundus image here
Supports PNG, JPG, JPEG, TIFF
Engine: Lesion Analytics & Groq AI

GAN Neural Restoration Comparison

Restored using model checkpoint checkpoints/best.pt

Checkpoint: best.pt (PSNR: 27.92 dB)
Restored Output
Degraded Input
↔
← Degraded Input Image best.pt Restored Output (with Vein Enhancement) →

Groq Vision Interactive ROI Inspector

Click anywhere on the retina photo to inspect a specific anatomical zone with Groq Vision AI.

ICDR Grade: Level 2
Click to inspect ROI
📍 Click on Fovea, Macula, Optic Disc, or Arcades to trigger Groq Vision AI inspection

📍 Click Any Region to Inspect

No region selected. Click on the retina image on the left to analyze that specific anatomical zone with Groq Vision AI (Llama 3.2 11B / 3.3 70B).

Quantitative Retinal Biomarkers

Vessel Tortuosity
1.142
Normal (1.05 - 1.20)
Fractal Dimension
1.528 D
Normal (1.40 - 1.70)
Arteriovenous Ratio
0.723
Zone B Standard
📷 Clinical Image Quality & Gradability Index (ISO 10940)
Evaluates sharpness, Laplacian variance, and illumination entropy for tele-ophthalmology AI grading.
GRADABLE (Optimal AI Clarity)

🎯 Retinal Lesion Quantification & Quadrant Breakdown

Automated lesion counting, pathology surface area percentage, and 4-quadrant spatial distribution.

Lesion Analytics Active
Total Lesions
14 Detected
Affected Retinal Area
2.4% Area
Hemorrhages
9 Focuses
Exudates / Micro-spots
5 Foci
Lesion Overlay

📍 4-Quadrant Pathology Distribution

Superior-Temporal Quadrant 6 Lesions (43%)
Inferior-Temporal Quadrant 5 Lesions (36%)
Superior-Nasal Quadrant 2 Lesions (14%)
Inferior-Nasal Quadrant 1 Lesion (7%)
💡 Clinical Guidance: Highest pathology density is localized in the Superior-Temporal Quadrant near the macula. Follow-up OCT recommended.

RetinaAI Diagnostic Clinical Summary

Report ID: #R-2026-8492 • Lesion Analytics & Groq Vision AI

Executive Summary: High-resolution neural GAN restoration (using model checkpoints/best.pt), Retinal Lesion Quantification, and Groq Vision AI clinical reasoning were performed on the uploaded fundus image.


Diagnostic Findings: Automated pathology evaluation identified non-proliferative changes. ICDR Severity Grade: Level 2 (Moderate Non-Proliferative Diabetic Retinopathy).


Biomarker Assessment: Vascular tortuosity index is 1.142 (normal range), fractal dimension is 1.528 D, and AVR ratio is 0.723.

💬 Groq AI Co-Pilot
Hello! I am your AI clinical assistant powered by Groq Llama-3.3-70b and best.pt. Ask me anything about retinal findings or biomarkers.