Representation Learning for Retinal Vasculature Embeddings. In: Cardoso MJ, Arbel T, Melbourne A, Bogunovic H, Moeskops P, Chen X, Schwartz E, Garvin M, Robinson E, Trucco E, Ebner M, Xu Y, Makropoulos A, Desjardin A, Vercauteren T, editors. Fetal, Infant and Ophthalmic Medical Image Analysis. Cham: Springer International Publishing; 2017. pp. 243-50. [DOI: 10.1007/978-3-319-67561-9_28]

Towards Stroke Biomarkers on Fundus Retinal Imaging: A Comparison Between Vasculature Embeddings and General Purpose Convolutional Neural Networks

Fundus Retinal imaging is an easy-to-acquire modality typically used for monitoring eye health. Current evidence indicates that the retina, and its vasculature in particular, is associated with other disease processes making it an ideal candidate for biomarker discovery. The development of these biomarkers has typically relied on predefined measurements, which makes the development process slow. Recently, representation learning algorithms such as general purpose convolutional neural networks or vasculature embeddings have been proposed as an approach to learn imaging biomarkers directly from the data, hence greatly speeding up their discovery. In this work, we compare and contrast different state-of-the-art retina biomarker discovery methods to identify signs of past stroke in the retinas of a curated patient cohort of 2,472 subjects from the UK Biobank dataset. We investigate two convolutional neural networks previously used in retina biomarker discovery and directly trained on the stroke outcome, and an extension of the vasculature embedding approach which infers its feature representation from the vasculature and combines the information of retinal images from both eyes.In our experiments, we show that the pipeline based on vasculature embeddings has comparable or better performance than other methods with a much more compact feature representation and ease of training.Clinical Relevance-This study compares and contrasts three retinal biomarker discovery strategies, using a curated dataset of subject evidence, for the analysis of the retina as a proxy in the assessment of clinical outcomes, such as stroke risk.

A multistep deep learning framework for the automated detection and segmentation of astrocytes in fluorescent images of brain tissue

While astrocytes have been traditionally described as passive supportive cells, studies during the last decade have shown they are active players in many aspects of CNS physiology and function both in normal and disease states. However, the precise mechanisms regulating astrocytes function and interactions within the CNS are still poorly understood. This knowledge gap is due in large part to the limitations of current image analysis tools that cannot process astrocyte images efficiently and to the lack of methods capable of quantifying their complex morphological characteristics. To provide an unbiased and accurate framework for the quantitative analysis of fluorescent images of astrocytes, we introduce a new automated image processing pipeline whose main novelties include an innovative module for cell detection based on multiscale directional filters and a segmentation routine that leverages deep learning and sparse representations to reduce the need of training data and improve performance. Extensive numerical tests show that our method performs very competitively with respect to state-of-the-art methods also in challenging images where astrocytes are clustered together. Our code is released open source and freely available to the scientific community.

Representation Learning of 3D Brain Angiograms, an Application for Cerebral Vasospasm Prediction

Stroke is the fifth leading cause of death in the United States. Subarachnoid hemorrhage (SAH) is a type of stroke often caused by the spontaneous rupture of a cerebral aneurysm. About 30% of the SAH patients develop delayed cerebral ischemia (DCI) a serious secondary complication with devastating impact. Cerebral vasospasm is one of the major precursors of DCI. Predicting the risk of vasospasm would enable better treatment and improved outcomes. Our overarching goal is to find a brain vasculature representation that can be used to find predictive image-based biomarkers. We propose a new methodology that leverages sparse dictionary learning and covariance-based features in order to encode the whole vessel structure in a vector of fixed size. Using 3D brain angiograms, we use this vasculature representation to train a logistic regression model to predict the occurrence of cerebral vasospasm with an area under the ROC curve of 0.93.

Deep Learning in Cardiology

The medical field is creating large amount of data that physicians are unable to decipher and use efficiently. Moreover, rule-based expert systems are inefficient in solving complicated medical tasks or for creating insights using big data. Deep learning has emerged as a more accurate and effective technology in a wide range of medical problems such as diagnosis, prediction, and intervention. Deep learning is a representation learning method that consists of layers that transform data nonlinearly, thus, revealing hierarchical relationships and structures. In this review, we survey deep learning application papers that use structured data, and signal and imaging modalities from cardiology. We discuss the advantages and limitations of applying deep learning in cardiology that also apply in medicine in general, while proposing certain directions as the most viable for clinical use.