A deep learning mobile-based image analysis for cervical cancer detection

This dataset provides a structured environment for cervical cancer image analysis using machine learning and deep learning. It includes four main experiments:

Risk factor prediction using classical ML models

ROI segmentation using U-Net

Lesion detection using Detectron2

Binary classification of images and Kappa evaluation

The experiments are implemented in Jupyter notebooks. All datasets are reduced 10% samples. The images used are publicly available from the Intel MobileODT dataset. Originally, the project included a private dataset (CAIME), but for privacy reasons, those images were removed and replaced with public samples. Both Intel/ and test/ folders now contain only public data. Segmentation masks (.tif) were also included where filenames matched.

This environment was originally executed in a Docker container with GPU support (NVIDIA QUADRO), but the reduced version can be tested on CPU.