Deep learning for medical applications faces many unique challenges. A major challenge is the large amount of labelled data for training, while working in a relatively data scarce environment. Active learning can be used to overcome the vast data need challenge. A second challenged faced is poor performance outside of a experimental setting, contrary to the high requirement for safety and robustness. In this paper, we present a novel framework for estimating uncertainty metrics and incorporating a similarity measure to improve active learning strategies. To showcase effectiveness, a medical image segmentation task was used as an exemplar. In addition to faster learning, robustness was also addressed through adversarial perturbations. Using epistemic uncertainty and our framework, we can cut number of annotations needed by 39% and by 54% using epistemic uncertainty and a similarity metric.

The thickness and appearance of retinal layers are essential markers for diagnosing and studying eye diseases. Despite the increasing availability of imaging devices to scan and store large amounts of data, analyzing retinal images and generating trial endpoints has remained a manual, error-prone, and time-consuming task. In particular, the lack of large amounts of high-quality labels for different diseases hinders the development of automated algorithms. Therefore, we have compiled 5016 pixel-wise manual labels for 1672 optical coherence tomography (OCT) scans featuring two different diseases as well as healthy subjects to help democratize the process of developing novel automatic techniques. We also collected 4698 bounding box annotations for a subset of 566 scans across 9 classes of disease biomarker. Due to variations in retinal morphology, intensity range, and changes in contrast and brightness, designing segmentation and detection methods that can generalize to different disease types is challenging. While machine learning-based methods can overcome these challenges, high-quality expert annotations are necessary for training. Publicly available annotated image datasets typically contain few images and/or only cover a single type of disease, and most are only annotated by a single grader. To address this gap, we present a comprehensive multi-grader and multi-disease dataset for training machine learning-based algorithms. The proposed dataset covers three subsets of scans (Age-related Macular Degeneration, Diabetic Macular Edema, and healthy) and annotations for two types of tasks (semantic segmentation and object detection).