In this project, my teammate Chijoke and I explore the application of Artificial Intelligence in medical diagnostics, focusing specifically on the segmentation of brain tumors from MRI scans. Our aim is to leverage AI to enhance the precision of medical imaging, improving diagnosis, treatment planning, and monitoring of brain tumors.
We developed an AI model using deep learning, specifically the U-Net architecture, to accurately identify and segment brain tumor tissues from normal brain tissues in MRI scans. The project not only aims to push the boundaries of AI in medical applications but also serves as an educational journey for us in the realm of neuro-oncology.
Figure 1: Distribution of positive and negative diagnoses in our dataset.
The accurate segmentation of brain tumors is crucial for formulating effective treatment plans and can significantly influence patient outcomes. Through this project, we hope to enhance the quality of care provided to patients by integrating cutting-edge technology into healthcare.
Figure 2: MRI scans and corresponding segmentation masks.
We utilized TensorFlow and PyTorch to implement our models, faced and overcame challenges related to data quality, and learned the importance of model interpretability. The use of U-Net architecture facilitated efficient and precise localization necessary for tumor segmentation.
Figure 3: Binary Cross-Entropy Loss and Intersection over Union scores on the validation set.
The datasets used were complex, requiring meticulous preprocessing and augmentation to ensure the model's robustness. Collaboration was key to our success, allowing us to overcome the challenges of data handling and model training.
Figure 4: Sample MRI images with predicted tumor segmentations.
This project has been a profound learning experience, enhancing our understanding of AI and deep learning and refining our problem-solving and innovation skills.
Figure 5: Additional MRI scans with their respective segmentation masks.
We invite you to follow our journey, learn from our experiences, and engage with us as we continue to contribute to AI in healthcare. This repository serves as a record of our work and a platform for further development and collaboration.
- Clone the repository to your local machine.
- Explore the Jupyter notebooks that detail our experiments and results.
- Review the code in the
srcdirectory to understand the implementation of the U-Net model. - Use the
datadirectory to access the MRI scans and masks used in our project. https://www.kaggle.com/datasets/mateuszbuda/lgg-mri-segmentation
For more information or to join our efforts, please contact us at [email/contact].