Journal of Biomedical Informatics最新文献

Medical Visual Question Answering (Med-VQA) is a critical multimodal task with the potential to address the scarcity and imbalance of medical resources. However, most existing studies overlook the limitations of the inconsistency in information density between medical images and text, as well as the long-tail distribution in datasets, which continue to make Med-VQA an open challenge. To overcome these issues, this study proposes a Language-Guided Progressive Fusion Network (LGPFN) with three key modules: Question-Guided Progressive Multimodal Fusion (QPMF), Language-Gate Mechanism (LGM), and Triple Semantic Feature Alignment (TriSFA). QPMF progressively guides the fusion of visual and textual features using both global and local question representations. LGM, a linguistic rule-based module, distinguishes between Closed-Ended (CE) and Open-Ended (OE) samples, directing the fused features to the appropriate classifiers. Finally, TriSFA captures the rich semantic information of OE answers and mine the underlying associations among fused features, predicted answers, and ground truths, aligning them in a ternary semantic feature space. The proposed LGPFN framework outperforms existing state-of-the-art models, achieving the best overall accuracies of 80.39%, 84.07%, 75.74%, and 70.60% on the VQA-RAD, SLAKE, PathVQA, and VQA-Med 2019 datasets, respectively. These results demonstrate the effectiveness and generalizability of the proposed model, underscoring its potential as a medical Artificial Intelligent (AI) agent that could benefit universal health coverage.

Objective: Data journeys are a way to describe and interrogate "the life of data" (Bates et al 2010). Thus far, they have been used to clarify the mobile nature of data by visualizing the pathways made by handling and moving data. We wanted to use the data journeys method (Eleftheriou et al. 2018) to compare different data journeys by noticing repetitions, patterns, and gaps.

Methods: We conducted qualitative interviews with 43 evaluators, implementers and administrators associated with 21 clinical and training programs, called "Enterprise-Wide Initiatives" (EWIs) that are part of a national health system in the United States. We used inductive and deductive coding to identify narratives of data journeys, and then we used the "swim lane" (Collar et al 2012) format to make data journey maps based on those narratives.

Results: Unlike the actors in Eleftheriou et al. (2018)'s work, who built a data infrastructure to manage clinical data, the actors in our study built data infrastructures to evaluate clinical data. We created and compared two data journey maps that helped us explore differences in data production and management. In tracing the pathways available to the data entity of interest, and the processes through which the actors interacted with it, we noticed how the same piece of information was made to work in different ways.

Conclusions: Researchers often must build a new data infrastructures to respond to the unique needs of their evaluation work. Differing abilities lead to differences in what programs can build, and consequently what kinds of evaluation work they can support. With the goal of straightforward comparisons across different programs, a more limited focus on quantitative values, and a better description of the data journeys used by the evaluation teams, might facilitate more nuanced assessments of the evidence of complex outcomes.

Objective: Persona validation is a challenging task, often relying on costly external validation methods. The aim of this study was the development of a novel method for Personas validation based on data already available during their creation.

Methods: A novel approach based on self-supervised machine learning (SSML) was proposed. A training-test split was performed (80 %-20 %), with the training set used for Personas development. The obtained labels were used as input for a 5-fold cross-validation grid search, resulting in 5 optimal different models. The "weak" ground truth for the test set was determined using the trained clustering model, and was compared with the prediction obtained by the majority voting of the optimal models. Performance evaluation was conducted by means of weighted accuracy, precision, recall and F1 score.

Results: The proposed method was evaluated on two very different healthcare datasets composed by questionnaires. The former was presented 1070 subjects, resulting in three unbalanced Personas (P0 n = 100; P1 n = 292; P2 n = 464). The latter included 176 subjects with three slightly unbalanced Personas. (P0 n = 58; P1 n = 32; P2 n = 50). The SSML approach resulted capable of correctly differentiating the clusters with high values of weighted accuracy (88.27 % and 94.12 %), precision (87.11 % and 92.83 %), recall (86.92 % and 91.67 %), and F1 score (86.92 % and 91.76 %).

Conclusions: The proposed method showed high capabilities in generalization beyond the training data, validating the Personas' capability of stratifying the characteristics of target populations. Additionally, this method significantly reduced the costs to validate Personas when compared to other methods in current literature.