The genomic status of gliomas is a critical biomarker for diagnosis and prognosis. Among these, the 1p/19q co-deletion is a hallmark of oligodendrogliomas, associated with better prognosis and treatment response compared to non-co-deleted cases. However, accurately predicting such markers remains challenging, as existing algorithms often overlook essential biological information, such as tumor microenvironment and shape features, which reflect tumor aggressiveness and growth patterns, thereby limiting genotyping accuracy. To address these challenges, we propose 3D-DPANet, a three-dimensional direction-perception attention network that integrates MRI imaging and domain-specific medical priors to predict 1p/19q co-deletion status. By employing a novel direction-perception attention mechanism, the network captures spatial dependencies and enhances feature representation using multi-modal MRI sequences, including CE-T1WI and T2-FLAIR. Integrating tumor microenvironment and shape features allows the model to effectively reflect tumors’ biological characteristics, improving interpretability and performance in molecular classification. We validated 3D-DPANet on a curated dataset of 412 glioma cases (237 with 1p/19q co-deletion, 175 without) using five-fold cross-validation, achieving an average accuracy of 82.03% (84.15%, 80.49%, 82.93%, 79.27%, 83.33%), significantly outperforming state-of-the-art methods. These results underscore the importance of incorporating medical priors and demonstrate the potential of 3D-DPANet to advance glioma genotyping through a more biologically meaningful approach.

Segmentation of the AVP using MRI provides important quantitative tools for analyzing AVP morphology and trajectory. However, due to the "X"-shaped structure of the AVP, its elongated morphology located deep within brain tissue, complex anatomical environment, and significant inter-individual morphological differences, traditional morphology- and atlas-based methods struggle to achieve precise segmentation of the AVP, often leading to segmentation failure and limiting clinical application. To address these challenges, deep learning-based AVP segmentation methods have been widely studied. Existing methods face challenges such as loss of texture details, structural discontinuities, and poor generalization, as they fail to accurately extract AVP structures from low-contrast regions and perform global modeling, while also struggling to capture potential periodic patterns in image data, resulting in poor performance on unseen data. In this paper, we propose a global periodic spatiotemporal aware AVP segmentation model named BiFxLSTM-UNet, which is capable of accurately segmenting AVP structures in complex brain environments. Specifically, our model consists of two components: a Fourier Fusion-driven bidirectional xLSTM visual backbone and a UNet decoder. First, the Fourier Fusion-driven bidirectional xLSTM visual backbone captures tight global spatiotemporal relationships between AVP slices, performs global periodic modeling, and learns the periodic patterns underlying the image data. Then, the UNet decoder generates AVP images that closely resemble manual annotations by medical professionals. Extensive comparative experiments demonstrate that, compared to existing methods, our approach shows competitive performance in both qualitative and quantitative evaluations of image quality.

Multiple sclerosis (MS) and neuromyelitis optica spectrum disorder (NMOSD) are rare demyelinating diseases of the central nervous system. Limited sample sizes pose significant challenges for traditional convolutional neural networks (CNNs) in learning lesion features, particularly in identifying critical regions relevant to disease prediction. Due to the distinct lesion patterns between MS and NMOSD, the anterior visual pathway (AVP) plays a crucial role in early diagnosis. However, these lesions often exhibit low contrast, making them difficult to detect using conventional methods, despite their clearer representation in the frequency domain. Few studies have incorporated AVP as prior knowledge into deep learning frameworks or addressed the differences in lesion characteristics across the frequency domain. To tackle these challenges, we propose a multi-task network, VAE-FreqNet, which jointly performs AVP segmentation and disease classification. First, a dynamic decoupling strategy based on discrete cosine transform (DCT) is introduced, where the HFDownsample and HFTUpsample modules preserve high-frequency and low-frequency details lost during downsampling and upsampling, thereby enhancing subtle frequency-domain features. Additionally, we design a frequency-domain hierarchical variational autoencoder (FHVAE) module that employs HVAE blocks for variational inference, fusing decoupled and original features to generate synthetic representations, thus alleviating data scarcity. Extensive experiments demonstrate that VAE-FreqNet significantly improves both classification and segmentation performance.

Hemorrhagic transformation (HT) is a time-sensitive severe complication of endovascular thrombectomy for patients with ischemic stroke, and there is an urgent need to develop deep learning models to assist doctors in making rapid preliminary diagnoses. Currently popular Transformer deep learning architectures, while superior in modeling global relationships compared to traditional CNN, it still faces quadratic complexity issues when handling long sequences of medical images due to its inherent attention mechanism. In contrast, computational complexity of the Mamba model-based method grows linearly. Based on these findings, we have developed a novel Mamba model that effectively captures long-range dependencies and the sequential relationships among slices in high-dimensional medical image sequences. We evaluated the proposed model on a multi-center dataset. Experimental results show that our method outperforms other classical architectures and current advanced methods, validating the effectiveness and generalizability of the model composed of the aforementioned modules.

Hemorrhagic transformation (HT) is a severe complication of acute ischemic stroke (AIS) that can lead to disability or death. Accurate and timely risk assessment of HT is essential for clinicians to design effective treatment strategies. Previous studies in HT prediction have largely relied on machine learning and radiomics, which demand extensive manual data preprocessing by physicians. While some HT prediction models based on convolutional neural network (CNN) have been developed, they are limited in their ability to capture the sequential relationships between image slices and often lack the focus on crucial information. This study collected non-contrast computed tomography (NCCT) images and clinical data from 512 AIS patients across six hospitals to create a multi-center dataset. Based on the dataset, we propose CM-HTNet, a novel deep learning framework designed to predict HT risk in AIS patients following intravenous thrombolysis (IVT). CM-HTNet mimics the clinical process of reviewing NCCT images by focusing on key slices and integrating information from adjacent slices. It leverages CNNs to extract features from each NCCT slice and utilizes the Selective State-Space Model (SSM) within the Mamba framework to model sequential relationships between slices while prioritizing features relevant to HT prediction. Additionally, CM-HTNet incorporates the Neighborhood Rough Set (KRS) algorithm for clinical feature selection and cross-attention mechanisms to integrate clinical and imaging data for multimodal HT risk prediction. In testing on an external dataset from independent centers, CM-HTNet achieved a prediction accuracy of 88.85% and an AUC of 95.17%, showcasing its strong performance and generalization capabilities.

3D-High Efficiency Video Coding (3D-HEVC), as an extension of HEVC in the realm of three-dimensional video, has brought significant coding performance improvements. However, traditional 3D video coding has faced many challenges such as compression distortion in texture and depth videos, as well as non-occlusion issues in Depth Image Based Rendering (DIBR) synthesis, which directly affected the visual quality of synthesized views. A Two-Stream Pyramid Attention Residual Network (TPARN) is proposed to achieve the quality enhancement of synthesized views. First of all, the Global Residual Attention (GRA) module and the Local Pyramid Attention (LPA) module are designed to extract global context information and intricate local texture details, which achieve a comprehensive scene understanding and preserve essential details across different scales. In addition, the Pyramid Attention Module (PAM) and skip connections are utilized to extract multiscale features, promoting seamless interaction among features. Experimental results demonstrate that the proposed method effectively reduces distortion caused by view synthesis, outperforming the latest methods in terms of performance.

Intelligent medicine is eager to automatically generate radiology reports to ease the tedious work of radiologists. Previous researches mainly focused on the text generation with encoder-decoder structure, while CNN networks for visual features ignored the long-range dependencies correlated with textual information. Besides, few studies exploit cross-modal mappings to promote radiology report generation. To alleviate the above problems, we propose a novel end-to-end radiology report generation model dubbed Self-Supervised dual-Stream Network (S3-Net). Specifically, a Dual-Stream Visual Feature Extractor (DSVFE) composed of ResNet and SwinTransformer is proposed to capture more abundant and effective visual features, where the former focuses on local response and the latter explores long-range dependencies. Then, we introduced the Fusion Alignment Module (FAM) to fuse the dual-stream visual features and facilitate alignment between visual features and text features. Furthermore, the Self-Supervised Learning with Mask(SSLM) is introduced to further enhance the visual feature representation ability. Experimental results on two mainstream radiology reporting datasets (IU X-ray and MIMIC-CXR) show that our proposed approach outperforms previous models in terms of language generation metrics.

Medical imaging is widely used in hospital clinical workflows. Assisting physicians in diagnosis by automatically generating reports from radiological images is an unmet clinical demand and requires urgent attention. However, this task suffers from two significant problems: 1) visual and textual data biases, and 2) the Transformer decoder makes no distinction between visual and non-visual words. We propose a novel multi-task approach combining natural language processing with machine learning techniques to meet this clinical need, i.e., creating fluent and accurate radiology reports. We name our system as Multi-modal Adaptive Transformer (MATNet), which consists of three key modules. First, Multi-Modal Encoder (MME) explores the relationship between radiology images and clinical notes. Second, Disease Classifier (DC) classifies the states of each disease topic and provides state-aware disease embeddings to alleviate visual data bias. Last, Adaptive Decoder (AD) dynamically measures the contribution of source signals and target signals when generating the next word. Based on our evaluations using benchmark IU-XRay and MIMIC-CXR datasets, the proposed MATNet outperformed previous state-of-the-art models on language fluency and clinical accuracy metrics such as BLEU scores.