Retrieval-augmented generation (RAG) is widely adopted for radiology report generation with medical vision-language models, leveraging external reports as linguistic references. However, existing RAG methods rely primarily on dense embedding similarity, which may retrieve reports that are semantically related yet clinically inconsistent with respect to presence or laterality constraints. Such inconsistencies are often propagated into generation, resulting in contradictory or unsupported findings. We propose an evidence-guided retrieval-augmented framework EviRAG that decomposes retrieval into structured and unstructured alignment levels. First, we induce structured clinical triplets from both query and database cases through targeted visual interrogation, projecting images into a shared evidence space. Triplet-level alignment enforces explicit agreement over presence and laterality variables, yielding a clinically admissible candidate set via structural ranking. Within this constrained space, we perform semantic alignment in a shared multimodal embedding space to capture nuanced descriptive correspondence. The top-ranked reports and query image are jointly fed into a medical vision-language model for report generation. Comprehensive experiments on radiology report generation benchmarks show that EviRAG substantially reduces clinical inconsistencies compared to strong medical vision-language baselines. The source code is available at https://github.com/liamgu06/EviRAG.

Large deformation is a key issue for deformable medical image registration. Decomposing a large deformation into several small deformations is an efficient solution. Current decomposition methods can be classed into iteration-based and non-iterative-based approaches. However, compared to iteration-based methods, non-iterative-based methods have faster inference times but registration accuracy gaps. To alleviate this limitation, we design a novel Non-iterative Pyramid Network (NIPNet). Firstly, Our Dual-domain Feature Extraction Module (DFEM) extracts global and local features in the frequency and spatial domains, respectively. Hence, the model considers global and local deformations. Secondly, a Multi-scale Localization Information Fusion Module (MIFM) is applied to fuse the localization information of adjacent scales to assist the current level in obtaining a more accurate deformation field. Finally, a Pyramid Self-distillation Loss (PDL) is introduced to improve the registration accuracy by treating the final deformation field as a teacher to guide the intermediate deformation field. By conducting intensive experiments on two typical 3D brain MRI datasets, we verify that the proposed NIPNet outperforms SOTA iterative-based methods and requires only a similar runtime as non-iterative approaches. Code is available at https://github.com/JXT210/NIPNet.

Entering text precisely by voice, users might encounter colloquial inserts, inappropriate wording, and recognition errors, which brings difficulties to voice editing. Users need to locate the errors and then correct them. In eyes-free scenarios, this select-modify mode brings a cognitive burden and a risk of error. This paper introduces neural networks and pre-trained models to understand users’ revision intention based on semantics, reducing the need for the information from users’ statements. We present two strategies. One is to remove the colloquial inserts automatically. The other is to allow users to edit by just speaking out the target words without having to say the context and the incorrect text. Accordingly, our approach can predict whether to insert or replace, the incorrect text to replace, and the position to insert. We implement these strategies in SmartEdit, an eyes-free voice input agent controlled with earphone buttons. The evaluation shows that our techniques reduce the cognitive load and decrease the average failure rate by 54.1% compared to descriptive command or re-speaking.

Timely access to professional medical advice is crucial for patient health outcomes. Traditional offline, one-on-one patient-provider interactions are time consuming and costly. As a result, online "Ask the doctor" (AtD) services have become increasingly popular, where patients and caregivers can obtain advice from medical professionals at a lower information and transaction cost. In this paper, we present an empirical study of Fenda, an innovative AtD platform recently introduced in China where patients and caregivers can consult a wide variety of healthcare professionals for a small fee. Using qualitative research methods, we analyzed how patients and caregivers interact with medical professionals on this platform, focusing on the nature of the questions asked and user strategies to optimize the usage of the platform. We further derived implications for designing better online AtD services connecting patients and caregivers to medical professionals.