Department of Image Processing

Images are an essential part of medical research, diagnosis, screening and monitoring of diseases. For example, the capillary non-perfusion lesions in images of retina can aid to understand the aetiology of malaria and the shape of optic disc can help to diagnose glaucoma. Often a clinician would manually evaluate these temporal or imaging data, however this is a complex task for a human observer, it is therefore expensive in terms of time and human training, it is subjective and there is a risk that potentially important information is being disregarded. A key to interpret the imaging and temporal data is to create and employ efficient quantitative approaches. This is a growing area of research in statistics, machine learning, computer vision and data science. We have developed spatial statistical modelling approaches to tackle this challenge. Our model is based on hierarchical statistical modelling, it involves spatial correlations and we use empirical Bayes to calculate posterior probability of the disease category. In retinal images we showed that retinal capillary-nonperfusion is more common in patients who died from malaria, and we demonstrated that the shape of the optic nerve can diagnose glaucoma with a supreme data efficiency and with a high accuracy comparable to machine learning approaches.*

* Two references that will be discussed at the talk:

Ian J. C. MacCormick, Yalin Zheng, Silvester Czanner, Yitian Zhao, Peter J. Diggle, Simon P. Harding & Gabriela Czanner, Spatial statistical modelling of capillary non-perfusion in the retina, Scientific Reports, volume 7, Article number: 16792 (2017)

http://www.nature.com/articles/s41598-017-16620-x?WT.ec_id=SREP-692-2017...

Ian J. C. MacCormick, Bryan M. Williams, Yalin Zheng, Kun Li, Baidaa Al-Bander, Silvester Czanner, Rob Cheeseman, Colin E. Willoughby, Emery N. Brown, George L. Spaeth, Gabriela Czanner, Accurate, fast, data efficient and interpretable glaucoma diagnosis with automated spatial analysis of the whole cup to disc profile, PLOSONE, 2019

https://journals.plos.org/plosone/article/comments?id=10.1371/journal.po...