OCT Analysis

As part of my work at the Institute for Neuro- and Bioinformatics, my task was to evaluate the usefulness of optical coherence tomography data in combination with machine learning.

At the time of the study, OCT was still a comparatively young and sparsely explored imaging technique. The project analysed eye scans from different participants to investigate potential long-term effects of smoking.

We used classical machine-learning methods such as Random Forests and SVMs as well as deep neural networks. The models were trained to detect and classify patterns in high-resolution OCT images.

The project reached a classification accuracy of 67%. Because the dataset was limited, the next step was to continue with a larger dataset to improve robustness and generalisability.

Technologies: PyTorch, Machine Learning, Data Transformation