by Francesco Cocciro and Michele Lunelli
The cell detection part (Michele Lunelli) of the tool has also a Streamlit interface:
https://github.com/michlun/Cervical_cancer_cell_detection
https://cancercell.streamlit.app
The project aims to develop a web application with a simple graphical interface for cervical cancer risk assessment and evaluation of Pap test images.
-
(Francesco Cocciro) Cervical Cancer Risk Prediction: Upon selecting the first button, a page will be loaded, prompting the user to fill in specific fields. This functionality allows users to predict their probability of developing cervical cancer based on the input values. The application will utilize a predictive model trained on relevant data to provide this probability. Additionally, the project incorporates a text generator that explains the results obtained. This generator will provide detailed explanations in natural language. By generating informative and interpretable explanations, users will have a better understanding of the underlying reasons behind the model's decisions, promoting trust and transparency in the application.
-
(Michele Lunelli) Cell Type Evaluation from Pap Slide Images: Selecting a second button will load a page that allows users to upload a microscope image, either of whole slides or single cells. The application will employ trained deep neural networks to analyze the image and determine if it contains cervical cancer cells or not. Single cells can also be selected from whole slide images for further evaulation. The model's output will provide insights into the presence or absence of tumor cells, assisting in early detection and diagnosis. To enhance the interpretability of the results, the project incorporates Explainable Artificial Intelligence (XAI) techniques. Specifically, Grad-CAM++ maps are generated to identify which parts of the input image are crucial in determining the outcome of the cell type.
Overall, this project aims to provide a user-friendly web application that empowers users to assess the risk of cervical cancer and offers the ability to analyze Pap slide images for potential tumor presence. By leveraging machine learning techniques and incorporating XAI and text generation, the application can contribute to early detection, promote timely medical intervention, and improve outcomes for individuals at risk of cervical cancer.
The models are trained and validated using the following datasets:
- Risk factors: https://archive.ics.uci.edu/dataset/383/cervical+cancer+risk+factors
- Microscope cell images: https://www.cs.uoi.gr/~marina/sipakmed.html
Please go in the app folder and run app.py with Flask: flask --app app run
The home page is then available at http://127.0.0.1:5000/index