Brain Tumor Classification & Segmentation Model

Stage 01 Classification: A VGG16-based architecture, pretrained on ImageNet, was fine-tuned with additional dense layers and dropout for tumor-type prediction. It also can easily represent spatial hierarchies of MRI images and thus, is highly suitable in transfer learning due to its uniform architecture with stacked convolutional layers. The last SoftMax layer delivers four possibilities of classes.

Stage 02 Segmentation: The tumor regions were detected using a YOLOv8n instance segmentation model trained on 10 epochs with an overlapping of the bounding box and class confidence scores.

Conditional Workflow: A fundamental conditional logic is used to decide whether or not to perform segmentation.  

The classification model uses the MRI images whereby the accurate region of interest (ROI) is extracted using a binary thresholding method to remove only the most useful features of analysis. The segmentation process will enable accurate tumor region segmentation, which plays a vital role in accurate diagnosis