R&D
Built by our founderAI-assisted lung-cancer detection in CT scans
A research project applying computer vision to identify and estimate the probability of four lung-carcinoma subtypes from CT scans — built as a decision-support tool for physicians.
- AI
- Computer vision
- Medical imaging
- CRISP-DM
Built with
- Python
- TensorFlow
- VGG16
- AWS EC2
The challenge
Lung-cancer diagnosis relies on specialist physicians reviewing CT scans — a process exposed to human error (inexperience, visual and cognitive fatigue, inter-observer variability) and hard to scale against rising imaging volumes.
The solution
We built and cleaned the dataset (expanded with data augmentation), then trained a computer-vision model that analyzes a CT scan and estimates the probability that it shows one of four carcinoma subtypes — squamous cell, adenocarcinoma, large cell and small cell — or none of them, and deployed it on AWS EC2.
Results
The neural network reached 88.68% validation accuracy — above the 85% bar set for usefulness — and was delivered as a web prototype, usability-tested with physicians. It is a research project, not a clinical-use device.