Abstract
Content-based image retrieval (CBIR) is a search technology that could aid medical diagnosis by retrieving and presenting earlier reported cases that are related to the one being diagnosed. To retrieve relevant cases, CBIR systems depend on supervised learning to map low-level image contents to high-level diagnostic concepts. However, the annotation by medical doctors for training and evaluation purposes is a difficult and time-consuming task, which restricts the supervised learning phase to specific CBIR problems of well-defined clinical applications. This paper proposes a new technique that automatically learns the similarity between the several exams from textual distances extracted from radiology reports, thereby successfully reducing the number of annotations needed. Our method first infers the relation between patients by using information retrieval techniques to determine the textual distances between patient radiology reports. These distances are subsequently used to supervise a metric learning algorithm, that transforms the image space accordingly to textual distances. CBIR systems with different image descriptions and different levels of medical annotations were evaluated, with and without supervision from textual distances, using a database of computer tomography scans of patients with interstitial lung diseases. The proposed method consistently improves CBIR mean average precision, with improvements that can reach 38%, and more marked gains for small annotation sets. Given the overall availability of radiology reports in picture archiving and communication systems, the proposed approach can be broadly applied to CBIR systems in different medical problems, and may facilitate the introduction of CBIR in clinical practice.

Abstract

Abstract
Given a retinal image, can we automatically determine whether it is of high quality (suitable for medical diagnosis)? Can we also explain our decision, pinpointing the region or regions that led to our decision? Images from human retinas are vital for the diagnosis of multiple health issues, like hypertension, diabetes, and Alzheimer's; low quality images may force the patient to come back again for a second scanning, wasting time and possibly delaying treatment. However, existing retinal image quality assessment methods are either black boxes without explanations of the results or depend heavily on feature engineering or on complex and error-prone anatomical structures' segmentation. Therefore, we propose EyeQual, that solves exactly this problem. EyeQual is novel, fast for inference, accurate and explainable, pinpointing low-quality regions on the image. We evaluated EyeQual on two real datasets where it achieved 100% accuracy taking just 36 milliseconds for each image.

Abstract

Abstract

Abstract
Background Changes in the retinal vessel caliber are associated with a variety of major diseases, namely diabetes, hypertension and atherosclerosis. The clinical assessment of these changes in fundus images is tiresome and prone to errors and thus automatic methods are desirable for objective and precise caliber measurement. However, the variability of blood vessel appearance, image quality and resolution make the development of these tools a non-trivial task. Metholodogy A method for the estimation of vessel caliber in eye fundus images via vessel cross-sectional intensity profile model fitting is herein proposed. First, the vessel centerlines are determined and individual segments are extracted and smoothed by spline approximation. Then, the corresponding cross-sectional intensity profiles are determined, post-processed and ultimately fitted by newly proposed parametric models. These models are based on Difference-of-Gaussians (DoG) curves modified through a multiplying line with varying inclination. With this, the proposed models can describe profile asymmetry, allowing a good adjustment to the most difficult profiles, namely those showing central light reflex. Finally, the parameters of the best-fit model are used to determine the vessel width using ensembles of bagged regression trees with random feature selection. Results and conclusions The performance of our approach is evaluated on the REVIEW public dataset by comparing the vessel cross-sectional profile fitting of the proposed modified DoG models with 7 and 8 parameters against a Hermite model with 6 parameters. Results on different goodness of fitness metrics indicate that our models are constantly better at fitting the vessel profiles. Furthermore, our width measurement algorithm achieves a precision close to the observers, outperforming state-of-the art methods, and retrieving the highest precision when evaluated using cross-validation. This high performance supports the robustness of the algorithm and validates its use in retinal vessel width measurement and possible integration in a system for retinal vasculature assessment.