Breast Cancer Detection

Early detection,
powered by AI.

Upload a histology slide or pick a reference from the BreaKHis dataset. Our deep-learning model classifies benign vs malignant tissue with calibrated confidence scoring.

94.6%
Model accuracy
< 2s
Inference time
50K+
Training images
Analyze an image
BreaKHis — SOB_B_A_14-22549AB · 40X
BENIGN · ADENOSIS · 40X
Benign No malignant markers detected
Confidence score94.6%
Run Analysis

Upload or pick a sample

Use your own histology image — or try a reference slide from the BreaKHis dataset, grouped by diagnosis type and magnification level.

Research & educational use only. Not a medical device. Results must not be used for clinical diagnosis or treatment. Always consult a qualified healthcare professional.
Image Analysis
BreaKHis dataset · 40X / 100X / 200X / 400X magnification
Drop image here, or browse
Accepts JPG, PNG — max 10 MB
—
—
Benign samples Benign
Loading…
Malignant samples Malignant
Loading…
—
—
—
Sample result
ResNet-50 128×128 input Research use only

Built on
rigorous science.

Every architectural and training decision was made to maximise clinical relevance and diagnostic reliability.

01
Deep CNN Architecture
ResNet-50 backbone fine-tuned on curated BreaKHis dataset images. Transfer learning maximises performance with limited labelled medical data.
02
Real-Time Inference
Sub-2-second predictions on CPU. No specialised GPU hardware required — deployable on any standard VM.
03
Calibrated Confidence
Temperature-scaled probabilities ensure scores are statistically meaningful, not just softmax artifacts.
04
Privacy by Design
Images processed in an ephemeral session, never persisted. No patient identifiers stored or logged.
05
Grad-CAM Explainability
Gradient-weighted Class Activation Maps highlight which image regions most influenced the model's prediction.
06
REST API Backend
FastAPI backend with structured routes for both direct file upload and URL-based sample prediction.
Technology

Built with

Python 3.10
TensorFlow / Keras
FastAPI
Uvicorn
ResNet-50
OpenCV
NumPy / Pandas
Scikit-learn
BreaKHis Dataset
Docker
Jinja2
Nginx + Debian