ExplainabilityTransparencyAccountabilitycs.CV
FG-CXR: A Radiologist-Aligned Gaze Dataset for Enhancing Interpretability in Chest X-Ray Report Generation
Trong Thang Pham, Ngoc-Vuong Ho, Nhat-Tan Bui, Thinh Phan, Patel Brijesh, Donald Adjeroh, Gianfranco Doretto, Anh Nguyen, Carol C. Wu, Hien Nguyen, Ngan Le
November 23, 2024
Asian Conference on Computer Vision
12 citations
1 influential
Key Takeaways
-
Existing AI systems for chest X-ray analysis lack transparency about which parts of images they examine: The paper found that current report-generation models cannot adequately explain which information in X-rays they use or how long they focus on specific areas, making it difficult for radiologists to trust or verify AI decisions.
-
A new dataset bridges the gap between AI interpretability and radiologist expertise: The FG-CXR dataset provides fine-grained alignment between where radiologists look (gaze attention) and what they write in reports, addressing a significant limitation in existing datasets that had "significant misalignment between gaze location and report content."
-
AI systems should mimic how radiologists actually work, not just generate text: The researchers developed Gen-XAI, a new model that explicitly constrains its outputs to match both radiologist gaze patterns and diagnostic transcripts, suggesting that interpretable AI should replicate the actual diagnostic process rather than treating it as a black-box problem.
-
Responsible AI in medical imaging requires alignment with human expert behavior: The research emphasizes that for radiologists to comprehend and trust AI-generated reports, these systems must demonstrate they're attending to the same visual information and following similar reasoning patterns as experienced clinicians.
Abstract
Developing an interpretable system for generating reports in chest X-ray (CXR) analysis is becoming increasingly crucial in Computer-aided Diagnosis (CAD) systems, enabling radiologists to comprehend the decisions made by these systems. Despite the growth of diverse datasets and methods focusing on report generation, there remains a notable gap in how closely these models' generated reports align with the interpretations of real radiologists. In this study, we tackle this challenge by initially introducing Fine-Grained CXR (FG-CXR) dataset, which provides fine-grained paired information between the captions generated by radiologists and the corresponding gaze attention heatmaps for each anatomy. Unlike existing datasets that include a raw sequence of gaze alongside a report, with significant misalignment between gaze location and report content, our FG-CXR dataset offers a more grained alignment between gaze attention and diagnosis transcript. Furthermore, our analysis reveals that simply applying black-box image captioning methods to generate reports cannot adequately explain which information in CXR is utilized and how long needs to attend to accurately generate reports. Consequently, we propose a novel explainable radiologist's attention generator network (Gen-XAI) that mimics the diagnosis process of radiologists, explicitly constraining its output to closely align with both radiologist's gaze attention and transcript. Finally, we perform extensive experiments to illustrate the effectiveness of our method. Our datasets and checkpoint is available at https://github.com/UARK-AICV/FG-CXR.
ACCV 2024
Paper Info
arXiv ID2411.15413
PublishedNovember 23, 2024
Citations
12
Influential
1
Relevance82%
VenueAsian Conference on Computer Vision
TypeJournalArticle, Conference
Categorycs.CV
Responsible AI Topics
ExplainabilityTransparencyAccountability
Links
Citation data from Semantic Scholar