Artificial Intelligence Image diagnostics (AIID) (2018-2021)
The field of artificial intelligence (AI) has been revolutionized in recent years with the emergence of deep machine learning. The combination of increased computational power and availability of large amounts of annotated data has allowed for the development of algorithms that have surpassed human level performance in a variety of tasks.
AI deep learning algorithms can learn to find patterns in data and distinguish between examples belonging to different categories or classes. This ability has a vast application potential not least within the medical field where diseases are diagnosed based on the presence of specific biological markers. Often, these ‘biomarkers’ are present in images in the form of e.g. malformed anatomical structures or diseased tissue. Diagnosing diseases in images requires training and experience which is often only found in highly specialized medical professionals. Recently, AI and deep learning has demonstrated its potential for detecting and diagnosing a number of diseases such as brain and breast cancers in MRI and X-ray scans as well as a number of vision related illnesses in different ophthalmologic image modalities at near, or above human expert level performance.
Figure 1: Illustration of deep learning architecture for diagnosis of diabetic retinopathy. The algorithm learns the parameters of a function, which can map the retinal input image to a diagnosis through a network of feature extraction layers and a final classification layer.
The artificial intelligence image diagnostics (AIID) project is a three year PhD-study funded by the by Odense University Hospital SDCO PhD Fund where the overarching aim is to utilize deep learning in the screening procedure for a specific disease called diabetic retinopathy (DR). DR is a potentially vision threatening micro-vascular complication of diabetes and is currently the leading cause of blindness among working aged adults in the western world. In collaboration with experts from the department of ophthalmology at Odense University Hospital (OUH), a deep learning based AI system will be developed according to the standards of care at OUH and Steno Diabetes Center Odense (SDCO) and tested with regards to clinical implementation at SDCO. The project also has a number of secondary objectives relating to the use of deep machine learning for assessing the risk of developing diabetes related cardiovascular complications based on biomarkers found in ophthalmologic images as well as developing a general framework for using AI in medical image diagnostics at OUH.