PhD thesis

I defended my thesis in AI segmentation on 3D image data nearly 12 years ago. The thesis explores techniques to make a computer understand images of the heart and its vessels in 3D MRI and CT.  These were the days before deep learning, and the use of hand-crafted features to detect structures of interest in 3D was quite common. I must say I did not feel content with the techniques I had developed, as I knew all along the way I was using hand-crafted features based on observations that may not necessarily generalise to all different shapes and sizes. 

I went into areas in my domain during my postdoctoral years, and they were spent working at british image research centres where I explored different problems such as representing 3D objects of interest as 2D flat maps, and also tissue classification using machine learning. 

But I came back to my original PhD problem, in later years of my postdoc, and together with a very bright colleague, I was finally able to solve it using a deep neural net. I was somewhat content by then, as I was finally not hard-coding features or hand-crafting them, but the neural net was able to generalise by 'learning' features from training images. 

 In these years, I spent a lot of time doing two things mostly:  

• Innovating, as my post was publicly and charity funded for making progress to improve our understanding of cardiac diseases using large-scale image analysis 
• Giving back to the research AI community, by collecting images, curating them and sharing images with the community to train their AI models. 

 Here is a link to my PhD thesis