X-ray imaging plays a crucial role in diagnostic medicine. Yet, a significant portion of the global population lacks access to this essential technology due to a shortage of trained radiologists. Eye-tracking data and deep learning models can enhance X-ray analysis by mapping expert focus areas, guiding automated anomaly detection, optimizing workflow efficiency, and bolstering training methods for novice radiologists. However, the literature shows contradictory results regarding the usefulness of eye-tracking data in deep-learning architectures for abnormality detection. We argue that these discrepancies between studies in the literature are due to (a) the way eye-tracking data is (or is not) processed, (b) the types of deep learning architectures chosen, and (c) the type of application that these architectures will have. We conducted a systematic literature review using PRISMA to address these contradicting results. We analyzed 60 studies that incorporated eye-tracking data in a deep-learning approach for different application goals in radiology. We performed a comparative analysis to understand if eye gaze data contains feature maps that can be useful under a deep learning approach and whether they can promote more interpretable predictions. To the best of our knowledge, this is the first survey in the area that performs a thorough investigation of eye gaze data processing techniques and their impacts in different deep learning architectures for applications such as error detection, classification, object detection, expertise level analysis, fatigue estimation and human attention prediction in medical imaging data. Our analysis resulted in two main contributions: (1) taxonomy that first divides the literature by task, enabling us to analyze the value eye movement can bring for each case and build guidelines regarding architectures and gaze processing techniques adequate for each application, and (2) an overall analysis of how eye gaze data can promote explainability in radiology.
This study investigates the impact of machine learning models on the generation of counterfactual explanations by conducting a benchmark evaluation over three different types of models: a decision tree (fully transparent, interpretable, white-box model), a random forest (semi-interpretable, grey-box model), and a neural network (fully opaque, black-box model). We tested the counterfactual generation process using four algorithms (DiCE, WatcherCF, prototype, and GrowingSpheresCF) in the literature in 25 different datasets. Our findings indicate that: (1) Different machine learning models have little impact on the generation of counterfactual explanations; (2) Counterfactual algorithms based uniquely on proximity loss functions are not actionable and will not provide meaningful explanations; (3) One cannot have meaningful evaluation results without guaranteeing plausibility in the counterfactual generation. Algorithms that do not consider plausibility in their internal mechanisms will lead to biased and unreliable conclusions if evaluated with the current state-of-the-art metrics; (4) A counterfactual inspection analysis is strongly recommended to ensure a robust examination of counterfactual explanations and the potential identification of biases.
X-ray images are vital in medical diagnostics, but their effectiveness is limited without clinical context. Radiologists often find chest X-rays insufficient for diagnosing underlying diseases, necessitating the integration of structured clinical features with radiology reports. To address this, we introduce DALL-M, a novel framework that enhances clinical datasets by generating contextual synthetic data. DALL-M augments structured patient data, including vital signs (e.g., heart rate, oxygen saturation), radiology findings (e.g., lesion presence), and demographic factors. It integrates this tabular data with contextual knowledge extracted from radiology reports and domain-specific resources (e.g., Radiopaedia, Wikipedia), ensuring clinical consistency and reliability. DALL-M follows a three-phase process: (i) clinical context storage, (ii) expert query generation, and (iii) context-aware feature augmentation. Using large language models (LLMs), it generates both contextual synthetic values for existing clinical features and entirely new, clinically relevant features. Applied to 799 cases from the MIMIC-IV dataset, DALL-M expanded the original 9 clinical features to 91. Empirical validation with machine learning models - including Decision Trees, Random Forests, XGBoost, and TabNET - demonstrated a 16.5% improvement in F1 score and a 25% increase in Precision and Recall. DALL-M bridges an important gap in clinical data augmentation by preserving data integrity while enhancing predictive modeling in healthcare. Our results show that integrating LLM-generated synthetic features significantly improves model performance, making DALL-M a scalable and practical approach for AI-driven medical diagnostics.
Integrating eye gaze data with chest X-ray images in deep learning (DL) has led to contradictory conclusions in the literature. Some authors assert that eye gaze data can enhance prediction accuracy, while others consider eye tracking irrelevant for predictive tasks. We argue that this disagreement lies in how researchers process eye-tracking data as most remain agnostic to the human component and apply the data directly to DL models without proper preprocessing. We present EyeXNet, a multimodal DL architecture that combines images and radiologists’ fixation masks to predict abnormality locations in chest X-rays. We focus on fixation maps during reporting moments as radiologists are more likely to focus on regions with abnormalities and provide more targeted regions to the predictive models. Our analysis compares radiologist fixations in both silent and reporting moments, revealing that more targeted and focused fixations occur during reporting. Our results show that integrating the fixation masks in a multimodal DL architecture outperformed the baseline model in five out of eight experiments regarding average Recall and six out of eight regarding average Precision. Incorporating fixation masks representing radiologists’ classification patterns in a multimodal DL architecture benefits lesion detection in chest X-ray (CXR) images, particularly when there is a strong correlation between fixation masks and generated proposal regions. This highlights the potential of leveraging fixation masks to enhance multimodal DL architectures for CXR image analysis. This work represents a first step towards human-centered DL, moving away from traditional data-driven and human-agnostic approaches.
My thesis consists of investigating how chest X-ray images, radiologists' eye movements and patients' clinical data can be used to teach a machine how radiologists read and classify images with the goal of creating human-centric AI architectures that can (1) capture radiologists' search behavioural patterns using their eye-movements in order to improve classification in DL systems, and (2) automatically detect lesions in medical images using clinical data and eye tracking data. Heterogeneous data sources such as chest X-rays, radiologists' eye movements, and patients' clinical data can contribute to novel multimodal DL architectures that, instead of learning directly from images' pixels, will learn human classification patterns encoded in both the eye movements of the images' regions and patients' medical history. In addition to a quantitative evaluation, I plan to conduct questionnaires with expert radiologists to understand the effectiveness of the proposed multimodal DL architecture.
This paper proposes a novel multimodal DL architecture incorporating medical images and eye-tracking data for abnormality detection in chest x-rays. Our results show that applying eye gaze data directly into DL architectures does not show superior predictive performance in abnormality detection chest X-rays. These results support other works in the literature and suggest that human-generated data, such as eye gaze, needs a more thorough investigation before being applied to DL architectures.