GazeRIB: Systematic Evaluation of Multi-Reader Gaze-Guided Chest X-ray Learning

Large scale vision language models have shown promise in automating chest Xray interpretation, yet their clinical utility remains limited by a gap between model outputs and radiologist reasoning. Most systems optimize for semantic information without emulating how experts visually examine medical images, often overlooking critical findings or diverging from established diagnostic workflows. Radiologists follow structured protocols (e.g., the ABCDEF approach) that ensure all clinically relevant regions are systematically examined, reducing missed findings and supporting reliable diagnostic reasoning. We introduce GazeX, a vision language model that leverages radiologists' eye tracking data as a behavioral prior to model expert diagnostic reasoning. By incorporating gaze trajectories and fixation patterns into pretraining, GazeX learns to follow the spatial and temporal structure of radiologist attention and integrates observations in a clinically meaningful sequence. Using a curated dataset of over 30,000 gaze key frames from five radiologists, we demonstrate that GazeX produces more accurate, interpretable, and expert consistent outputs across radiology report generation, disease grounding, and visual question answering, utilizing 231,835 radiographic studies, 780,014 question answer pairs, and 1,162 image sentence pairs with bounding boxes. Unlike autonomous reporting systems, GazeX produces verifiable evidence artifacts, including inspection trajectories and finding linked localized regions, enabling efficient human verification and safe human AI collaboration. Learning through expert eyes provides a practical route toward more trustworthy, explainable, and diagnostically robust AI systems for radiology and beyond.

Expert eye movements provide a rich, passive source of domain knowledge in radiology, offering a powerful cue for integrating diagnostic reasoning into computer-aided analysis. However, direct integration into CNN-based systems, which historically have dominated the medical image analysis domain, is challenging: gaze recordings are sequential, temporally dense yet spatially sparse, noisy, and variable across experts. As a consequence, most existing image-based models utilize reduced representations such as heatmaps. In contrast, gaze naturally aligns with transformer architectures, as both are sequential in nature and rely on attention to highlight relevant input regions. In this work, we introduce FixationFormer, a transformer-based architecture that represents expert gaze trajectories as sequences of tokens, thereby preserving their temporal and spatial structure. By modeling gaze sequences jointly with image features, our approach addresses sparsity and variability in gaze data while enabling a more direct and fine-grained integration of expert diagnostic cues through explicit cross-attention between the image and gaze token sequences. We evaluate our method on three publicly available benchmark chest X-ray datasets and demonstrate that it achieves state-of-the-art classification performance, highlighting the value of representing gaze as a sequence in transformer-based medical image analysis.

Chest radiograph datasets often combine multiple images with Clinical Indication, Findings, and Impression, although these inputs are produced at different stages of care. We evaluated 15,000 ReXGradient-160K studies with two readable images and five CheXbert-derived report observations. Frozen DenseNet-121 and Bio+ClinicalBERT encoders were used to compare image-only, Indication-only, fixed-order multimodal, random-swap, DeepSets, and SectionGuard-MI models. Findings and Impression were evaluated only as post-hoc leakage controls. Models were trained with five seeds, and public-test uncertainty was estimated with 2,000 patient-cluster bootstrap replicates. Under U-Ones, macro AUROC was 0.643 for the primary image, 0.694 for two images, 0.749 for Indication, and 0.780 for ordinary two-image-plus-Indication fusion. SectionGuard-MI achieved AUROC 0.783 and AUPRC 0.260. Relative to ordinary fusion, its paired AUROC difference was 0.0031 (95% CI, -0.0042 to 0.0104; adjusted p=0.374), while its AUPRC difference was 0.0289 (95% CI, 0.0095 to 0.0413; adjusted p=0.004). DeepSets had the highest prospective AUROC point estimate (0.787), and random-swap fusion had the highest prospective AUPRC point estimate (0.265) with better calibration than SectionGuard-MI. Full report text alone reached AUROC 0.979 and AUPRC 0.836; AUROC remained above 0.973 after exact or expanded masking. These results show that prospective Indication is strongly associated with report-derived targets, permutation-aware fusion is competitive, and post-hoc report text creates substantial report-label circularity.