Developer
Brain MRI tumour ensemble
Combining convolutional networks to explore MRI tumour classification, calibration and model attention.
Overview
An experimental image-classification pipeline combining VGG16, DenseNet and EfficientNet backbones. The work examines both predictions and the image regions highlighted by Grad-CAM.
The question
A classifier score alone does not explain prediction confidence or which image regions influence a result.
My approach
Compare transfer-learning models, combine predictions, examine calibration and visualise spatial attention.
Outcome and limits
Model comparison, probability calibration and Grad-CAM visualisations.
An experimental image-classification project, not a clinically validated diagnostic system. Performance on new datasets and clinical settings would require separate evaluation.
Interested in this work? Get in touch.