A deep learning mobile-based image analysis for cervical cancer detection
Description
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.
Files
Steps to reproduce
1. It is recommended to create and activate a virtual environment to isolate dependencies. 2. Extract the ZIP archive. 3. Navigate to the extracted folder named 'colpotool-lab'. 4. Install the dependencies using: pip install -r requirements.txt 5. Launch Jupyter using: jupyter notebook 6. Open and run the notebooks located in the 'experiments/' directory. Note: The original experiments were executed in a Docker environment using a TITAN GPU. However, this reduced version can be executed on CPU for testing purposes.
Institutions
Categories
Funding
Universidad Técnica Particular de Loja