Kidney CT Scan
Tumor Classifier
A deep learning system built to help detect kidney tumors from CT scan images. Upload a scan and the model will tell you within seconds whether the kidney appears normal or shows signs of a tumor. Built with transfer learning, full experiment tracking, and a reproducible MLOps pipeline.
Drop your CT scan image here
or click to choose a file
Why VGG16?
VGG16 was chosen because its deep stack of simple 3x3 convolution layers is remarkably good at learning fine-grained textures, which is exactly what you need when distinguishing healthy renal tissue from abnormal cell growth in a CT scan. Pre-trained on ImageNet, its weights already encode a rich understanding of edges, shapes, and spatial patterns, making it an ideal starting point for medical imaging tasks where labelled data is limited.
How the Model Was Built
The training process used transfer learning. The VGG16 base layers were frozen to preserve the knowledge captured from ImageNet, and a custom classification head was added and fine-tuned on kidney CT scan images split 70 percent for training and 30 percent for validation. Every experiment was tracked end to end with MLflow on DagsHub, capturing parameters, metrics, and model artifacts for full auditability and comparison across runs.
MLOps Pipeline
The project is structured around four fully automated DVC pipeline stages: data ingestion, base model preparation, training, and evaluation. Each stage is versioned independently so that only what has changed is re-executed on the next run. Model metrics are pushed automatically to the MLflow registry, enabling side-by-side comparison of runs and straightforward model promotion to production.
Tech Stack
Built with tools that are standard in modern ML engineering teams.