Reports3D

A deep learning CNN solution which segments brain tumours from MRI scans and reconstructs with accurate location, size and orientation a 3D model of the tumour embedded in the patient brain, available for viewing in an Augmented Reality enviroment.

There exists a vast semantic gap between doctors and patients when it comes to medical imaging. This application is aimed to reduce this communication barrier by making extant tumours easier to visualise in 3 dimensions.

Architectue of model: U-NET Model

Image segmentation results:

Comparison of ground truth with predicted images

Performance:

Mean Dice Similarity Coefficient	Median Dice Similarity Coefficient
0.87	0.90

Final output:

See it in action here