Analyzing scan
Preprocessing image...

Deep Learning · Medical Image Analysis

Multi-Disease
Detection

An AI-powered diagnostic system analyzing chest X-rays, retinal fundus, brain MRI, and skin lesion images using DenseNet-121 and EfficientNet-B3 with Grad-CAM explainability.

Start Analysis How it works
4 Scan Modalities
25+ Disease Classes
0.85+ Target AUC
XAI Grad-CAM Ready

Detection Modules

Four imaging modalities

Chest X-Ray
Multi-label detection of 14 thoracic diseases from frontal chest radiographs using DenseNet-121.
DenseNet-121 Multi-label 14 diseases AUC > 0.85
Retinal Fundus
Diabetic retinopathy grading (0–4) from retinal fundus images using EfficientNet-B3.
EfficientNet-B3 DR Grading 5 stages Kappa > 0.80
Brain MRI
Tumor classification into glioma, meningioma, pituitary, or no tumor from MRI scans.
EfficientNet-B3 4 classes Grad-CAM Acc > 90%
Skin Lesion
Classification of 9 skin lesion types including melanoma detection from dermoscopic images.
EfficientNet-B3 9 classes ISIC 2019 Acc > 80%

Analysis Engine

Upload & Diagnose

Select scan type

Drop medical image here
PNG, JPG, DICOM · Max 10MB
Preview
Ready for analysis

Upload a medical image
and run analysis to see results

BRAIN MRI · TUMOR CLASSIFICATION Analysis complete
⚠ For research and educational purposes only. Not a substitute for clinical diagnosis.
Grad-CAM · Explainable AI Highlighted regions show areas the model focused on for its prediction
Original
Original scan
Heatmap
Grad-CAM heatmap
Overlay
Overlay
Low attention
High attention

Methodology

How MediScan works

1
Image Upload
Medical image uploaded and validated. Format and modality detected automatically.
2
Preprocessing
Resize, normalize using ImageNet statistics, apply modality-specific enhancements.
3
Deep Learning
DenseNet-121 or EfficientNet-B3 extracts features and outputs disease probabilities.
4
Grad-CAM
Gradient-weighted class activation maps highlight the regions that drove the prediction.
5
Report
Probability scores, diagnosis summary, and visual explanation presented to the user.

Technology Stack

Built with precision

Model · Chest
DenseNet-121
Pretrained ImageNet · 14-output sigmoid · BCEWithLogitsLoss
Model · Retina / Brain / Skin
EfficientNet-B3
Pretrained ImageNet · Custom classifier head · CrossEntropyLoss
Framework
PyTorch 2.x
CUDA 12.1 · RTX 3050 · AdamW + CosineAnnealing
Explainability
Grad-CAM
Gradient-weighted class activation maps · Last conv block
Backend
FastAPI
REST API · Model serving · Image preprocessing pipeline
Frontend
HTML · CSS · JS
Vanilla stack · No framework · Medical-grade dark UI