Timmothy Dangeon, a student in DSTI’s MSc in Data Science & AI, has developed a deep learning solution for the early detection of brain tumours. Using the EfficientNetB0 model, this project demonstrates how AI can enhance the accuracy and efficiency of diagnosing brain tumours, including glioma, meningioma, and pituitary tumours, from MRI scans.
Early and accurate detection of brain tumours is crucial for improving patient outcomes. This project tackles this challenge by automating the detection process using AI, reducing the potential for human error and accelerating diagnosis.
As the healthcare industry increasingly turns to AI for diagnostic tools, this project stands out as a prime example of how deep learning can be applied to medical imaging, offering a solution that is both timely and necessary in today’s medical landscape.
The project is built using Python, a preferred language in Data science and AI. Timmothy employed TensorFlow and Keras for model development, OpenCV for image processing, and FastAPI to create an API interface. Streamlit was used to build a user-friendly web application, making the model accessible to users for testing.
The project utilised a dataset of 7,023 MRI scans, including T1-weighted, T2-weighted, and FLAIR images, sourced from Kaggle. This comprehensive dataset was selected for its variety and volume, providing a solid foundation for training the model to distinguish between different types of brain tumours and healthy tissues.
Timmothy chose the EfficientNetB0 model for its efficiency and accuracy in image classification tasks. By applying transfer learning, he fine-tuned the pre-trained model to specialize in brain tumour detection, significantly reducing the computational resources required while achieving high performance.
MRI images were pre-processed to remove duplicates and crop unnecessary borders, ensuring that the data was clean and standardized for input into the model.
The EfficientNetB0 model was trained on the processed dataset, with checkpoints in place to save the best-performing version based on validation metrics.
Fast API was used to manage the backend operations, while Streamlit was employed to create an intuitive front-end interface. This web application allows users to upload MRI images and receive predictions from the model.
The final model was deployed on a web platform, where it can be accessed and tested in real-time, showcasing the project’s practical application.
Balancing the dataset to avoid bias towards certain tumour types was a significant challenge. Timmothy used oversampling techniques to ensure that the model was exposed to a balanced representation of all tumour classes during training.
The model achieved an accuracy of 0.997, a sensitivity of 0.997, and a specificity of 0.999 on the test dataset, demonstrating its reliability in identifying both tumour-positive and tumour-negative cases. Techniques like Grad-CAM were used to visualize the regions of the brain that influenced the model’s predictions, providing additional validation for its accuracy.
This project has the potential to significantly impact the healthcare industry by providing a tool for the early detection of brain tumours. Its open-source nature encourages further research and development, making it a valuable contribution to the field of medical AI.
Timmothy reflected, “This project has allowed me to combine my pharmaceutical background with AI, creating something that can truly make a difference in healthcare. It’s been a rewarding experience to see how technology can be used to solve real-world problems.”
Through this project, Timmothy gained expertise in deep learning, particularly in using TensorFlow and Keras for model development. He also developed skills in data preprocessing and web app deployment, making him proficient in end-to-end project implementation.
The skills acquired through this project are highly sought after in both the AI and healthcare industries. As demand for AI-driven diagnostic tools grows, Timmothy’s experience positions him as a strong candidate for roles in these fields.
Interested in AI and healthcare? Learn more about DSTI’s Applied MSc in Data Science & AI and see how you can work on groundbreaking projects like this.
At DSTI School of Engineering, our mission has always been to bridge the gap between academic excellence and industry-driven innovation. This month, our CEO and Dean, Sébastien Corniglion, embarked on a high-impact tour across India to solidify international partnerships and create direct pathways for the next generation of data scientists
Explore the dynamic roles of AI Engineers in industrialising AI, contrasted with Data Scientists, in our comprehensive guide. Learn how DSTI’s Applied MSc programmes uniquely equip you for a successful career as an AI Engineer. Discover skills, collaboration, and career prospects in AI engineering.