This project applies deep learning to a high-stakes medical imaging problem: the automatic detection of brain tumors from MRI scans. By combining rigorous dataset preprocessing, transfer learning with VGG16, and a React-based user interface, the system delivers a practical tool that bridges the gap between model training and clinical usability.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/JuanRojasDev/juan-rojas-portafolio-web/llms.txt
Use this file to discover all available pages before exploring further.
Project overview
Date: January 2023 – March 2023 The Brain Tumor Detection application was built end-to-end — from dataset preparation to a deployed web interface. The workflow covered dataset augmentation to improve generalization, model training using the VGG16 convolutional architecture, a Python API serving model predictions, and a React front-end for user interaction.The model achieved 99.2% accuracy in detecting brain tumors from medical MRI images.
Dataset augmentation and model training
The training dataset was preprocessed and augmented to reduce overfitting and improve accuracy on unseen images. Augmentation techniques expanded the effective size of the training set, allowing the model to generalize better to real-world MRI variations. Transfer learning was applied using the VGG16 architecture pretrained on ImageNet. Fine-tuning the final layers on the medical imaging dataset allowed the model to leverage deep visual feature extraction while adapting to the specific classification task. The trained model weights were serialized using Pickle for efficient loading at inference time.Tech stack
| Component | Technology |
|---|---|
| Model architecture | VGG16 (transfer learning) |
| Deep learning framework | TensorFlow, Keras |
| Model serialization | Pickle |
| Backend API | Python |
| Front-end interface | React |