Weakly-Supervised Learning Significantly Reduces the Number of Labels Required for Intracranial Hemorrhage Detection on Head CT

Modern machine learning pipelines, in particular those based on deep learning (DL) models, require large amounts of labeled data. For classification problems, the most common learning paradigm consists of presenting labeled examples during training, thus providing \emph{strong supervision} by directly presenting examples from the different classes, e.g. positive and negative samples. As a result, the adequate training of these models demands the curation of large datasets with high-quality labels. This constitutes a major obstacle for the development of DL models in radiology---in particular for cross-sectional imaging (e.g., computed tomography [CT] scans)---where labels must come from manual annotations by expert radiologists at the image or slice-level (as opposed to the examination level, such as could be obtained using natural language processing of radiology reports). 
This work studies the question of what kind of labels should be collected for the problem of intracranial hemorrhage detection in brain CT. We investigate whether image-level annotations should be preferred to examination-level ones. By framing this task as a Multiple Instance Learning (MIL) problem, and employing modern attention-based DL architectures, we analyze the degree to which different levels of supervision improve the detection performance. We find that strong supervision (learning with local image-level annotations) and weak supervision (learning with only global examination-level labels) achieve comparable performance in both examination- and image-level hemorrhage detection, as well as in hemorrhage localization at the pixel-level (explainability). Furthermore, we study this behavior as a function of the number of labels available during training. Our results suggest that local labels may not be necessary at all, drastically reducing the time and cost involved in collecting and curating datasets.