Improving FTIR Spectroscopy-based Prostate Cancer Classification through Deep Learning.

Project objectives:

Investigate the image processing and neural network methods that work best to perform various machine learning tasks on FTIR prostate cancer data.

Investigate the explainability of any developed techniques to aid reliability, increase trust from pathologists and clinicians, and increase chance of adoption in a clinical setting.


Project approaches:

First literature is planned, particularly in areas of prostate cancer diagnosis and chemistry, and previous applications of machine learning to FTIR data.

Then ongoing experimentation with different types of deep neural networks and image processing techniques will be important to find methods to pursue further in research. Different machine learning tasks have different difficulties. Grading cancer will always be hard task compared to differentiating cancerous and non cancerous samples. Spatial information will be explored through the use of, but no limited to, the Markov random field.

Collaborations with chemical engineering colleagues in FTIR and pathologists will be important and helpful for gradually gaining more insight to the problem and better understanding of the data and its meaning, while processing the data is a large undertaking.