Chest X-rays (CXRs) are the most frequently performed imaging test in clinical practice. Recent advances in the development of vision-language foundation models (FMs) give rise to the possibility of performing automated CXR interpretation, which can assist physicians with clinical decision-making and improve patient outcomes. However, developing FMs that can accurately interpret CXRs is challenging due to the (1) limited availability of large-scale vision-language datasets in the medical image domain, (2) lack of vision and language encoders that can capture the complexities of medical data, and (3) absence of evaluation frameworks for benchmarking the abilities of FMs on CXR interpretation. In this work, we address these challenges by first introducing \emph{CheXinstruct} - a large-scale instruction-tuning dataset curated from 28 publicly-available datasets. We then present \emph{CheXagent} - an instruction-tuned FM capable of analyzing and summarizing CXRs. To build CheXagent, we design a clinical large language model (LLM) for parsing radiology reports, a vision encoder for representing CXR images, and a network to bridge the vision and language modalities. Finally, we introduce \emph{CheXbench} - a novel benchmark designed to systematically evaluate FMs across 8 clinically-relevant CXR interpretation tasks. Extensive quantitative evaluations and qualitative reviews with five expert radiologists demonstrate that CheXagent outperforms previously-developed general- and medical-domain FMs on CheXbench tasks. Furthermore, in an effort to improve model transparency, we perform a fairness evaluation across factors of sex, race and age to highlight potential performance disparities. Our project is at \url{https://stanford-aimi.github.io/chexagent.html}.
Vision-language models (VLMs), such as CLIP and ALIGN, are generally trained on datasets consisting of image-caption pairs obtained from the web. However, real-world multimodal datasets, such as healthcare data, are significantly more complex: each image (e.g. X-ray) is often paired with text (e.g. physician report) that describes many distinct attributes occurring in fine-grained regions of the image. We refer to these samples as exhibiting high pairwise complexity, since each image-text pair can be decomposed into a large number of region-attribute pairings. The extent to which VLMs can capture fine-grained relationships between image regions and textual attributes when trained on such data has not been previously evaluated. The first key contribution of this work is to demonstrate through systematic evaluations that as the pairwise complexity of the training dataset increases, standard VLMs struggle to learn region-attribute relationships, exhibiting performance degradations of up to 37% on retrieval tasks. In order to address this issue, we introduce ViLLA as our second key contribution. ViLLA, which is trained to capture fine-grained region-attribute relationships from complex datasets, involves two components: (a) a lightweight, self-supervised mapping model to decompose image-text samples into region-attribute pairs, and (b) a contrastive VLM to learn representations from generated region-attribute pairs. We demonstrate with experiments across four domains (synthetic, product, medical, and natural images) that ViLLA outperforms comparable VLMs on fine-grained reasoning tasks, such as zero-shot object detection (up to 3.6 AP50 points on COCO and 0.6 mAP points on LVIS) and retrieval (up to 14.2 R-Precision points).
Radiology report summarization is a growing area of research. Given the Findings and/or Background sections of a radiology report, the goal is to generate a summary (called an Impression section) that highlights the key observations and conclusions of the radiology study. Recent efforts have released systems that achieve promising performance as measured by widely used summarization metrics such as BLEU and ROUGE. However, the research area of radiology report summarization currently faces important limitations. First, most of the results are reported on private datasets. This limitation prevents the ability to reproduce results and fairly compare different systems and solutions. Secondly, to the best of our knowledge, most research is carried out on chest X-rays. Sometimes, studies even omit to mention the concerned modality and anatomy in the radiology reports used for their experiments. To palliate these limitations, we propose a new dataset of six different modalities and anatomies based on the MIMIC-III database. We further release our results and the data splits used to carry out our experiments. Finally, we propose a simple report summarization system that outperforms the previous replicable research on the existing dataset.
Machine learning models that achieve high overall accuracy often make systematic errors on important subsets (or slices) of data. Identifying underperforming slices is particularly challenging when working with high-dimensional inputs (e.g. images, audio), where important slices are often unlabeled. In order to address this issue, recent studies have proposed automated slice discovery methods (SDMs), which leverage learned model representations to mine input data for slices on which a model performs poorly. To be useful to a practitioner, these methods must identify slices that are both underperforming and coherent (i.e. united by a human-understandable concept). However, no quantitative evaluation framework currently exists for rigorously assessing SDMs with respect to these criteria. Additionally, prior qualitative evaluations have shown that SDMs often identify slices that are incoherent. In this work, we address these challenges by first designing a principled evaluation framework that enables a quantitative comparison of SDMs across 1,235 slice discovery settings in three input domains (natural images, medical images, and time-series data). Then, motivated by the recent development of powerful cross-modal representation learning approaches, we present Domino, an SDM that leverages cross-modal embeddings and a novel error-aware mixture model to discover and describe coherent slices. We find that Domino accurately identifies 36% of the 1,235 slices in our framework - a 12 percentage point improvement over prior methods. Further, Domino is the first SDM that can provide natural language descriptions of identified slices, correctly generating the exact name of the slice in 35% of settings.
Named entity disambiguation (NED), which involves mapping textual mentions to structured entities, is particularly challenging in the medical domain due to the presence of rare entities. Existing approaches are limited by the presence of coarse-grained structural resources in biomedical knowledge bases as well as the use of training datasets that provide low coverage over uncommon resources. In this work, we address these issues by proposing a cross-domain data integration method that transfers structural knowledge from a general text knowledge base to the medical domain. We utilize our integration scheme to augment structural resources and generate a large biomedical NED dataset for pretraining. Our pretrained model with injected structural knowledge achieves state-of-the-art performance on two benchmark medical NED datasets: MedMentions and BC5CDR. Furthermore, we improve disambiguation of rare entities by up to 57 accuracy points.
Automated emotion classification could aid those who struggle to recognize emotion, including children with developmental behavioral conditions such as autism. However, most computer vision emotion models are trained on adult affect and therefore underperform on child faces. In this study, we designed a strategy to gamify the collection and the labeling of child affect data in an effort to boost the performance of automatic child emotion detection to a level closer to what will be needed for translational digital healthcare. We leveraged our therapeutic smartphone game, GuessWhat, which was designed in large part for children with developmental and behavioral conditions, to gamify the secure collection of video data of children expressing a variety of emotions prompted by the game. Through a secure web interface gamifying the human labeling effort, we gathered and labeled 2,155 videos, 39,968 emotion frames, and 106,001 labels on all images. With this drastically expanded pediatric emotion centric database (>30x larger than existing public pediatric affect datasets), we trained a pediatric emotion classification convolutional neural network (CNN) classifier of happy, sad, surprised, fearful, angry, disgust, and neutral expressions in children. The classifier achieved 66.9% balanced accuracy and 67.4% F1-score on the entirety of CAFE as well as 79.1% balanced accuracy and 78.0% F1-score on CAFE Subset A, a subset containing at least 60% human agreement on emotions labels. This performance is at least 10% higher than all previously published classifiers, the best of which reached 56.% balanced accuracy even when combining "anger" and "disgust" into a single class. This work validates that mobile games designed for pediatric therapies can generate high volumes of domain-relevant datasets to train state of the art classifiers to perform tasks highly relevant to precision health efforts.
Datasets extracted from social networks and online forums are often prone to the pitfalls of natural language, namely the presence of unstructured and noisy data. In this work, we seek to enable the collection of high-quality question-answer datasets from social media by proposing a novel task for automated quality analysis and data cleaning: question-answer (QA) plausibility. Given a machine or user-generated question and a crowd-sourced response from a social media user, we determine if the question and response are valid; if so, we identify the answer within the free-form response. We design BERT-based models to perform the QA plausibility task, and we evaluate the ability of our models to generate a clean, usable question-answer dataset. Our highest-performing approach consists of a single-task model which determines the plausibility of the question, followed by a multi-task model which evaluates the plausibility of the response as well as extracts answers (Question Plausibility AUROC=0.75, Response Plausibility AUROC=0.78, Answer Extraction F1=0.665).