Accurate prediction of treatment response and selection of optimal treatments remain challenging in epilepsy management. With no reliable surrogate biomarkers for treatment response, the current process of selecting an antiseizure medication remains largely a trial-and-error approach. Other non-pharmacological treatment options, such as epilepsy surgery, are viable alternatives for patients with drug-resistant epilepsy. Statistical and machine learning techniques have been used to predict seizure outcomes associated with antiseizure medications and epilepsy surgery. Recent breakthroughs in deep learning have unveiled new pathways and opportunities, potentially revolutionising personalised treatment selection in health care. In this Review, we explore a broad range of studies that have used various statistical and machine learning methodologies, with particular emphasis on state-of-the-art deep learning techniques to predict the outcomes of both pharmaceutical and surgical treatments for epilepsy. We also review potential future research trajectories and address the inherent challenges of incorporating machine learning into the clinical management of epilepsy.
Parahoric Lusztig induction gives a broad class of virtual smooth representations of parahoric subgroups in a p-adic group, serving as a natural generalization of classical Lusztig induction. This construction has useful applications in the representation theory of p-adic groups. In this paper, we prove a scalar product formula for parahoric Lusztig induction under the (L, G)-generic condition, which can be viewed as a parahoric analogue of a classical result of Lusztig in 1976. As an application, we describe the irreducible decomposition of the parahoric Deligne-Lusztig representation in the case of elliptic tori. (c) 2026 Elsevier Inc. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
Complex bone defects remain major clinical challenges due to their multifaceted pathological microenvironment. In recent years, hydrogen (H2) therapy has emerged as a rapidly developing strategy for bone repair, leveraging its redox-regulatory and other microenvironment-modulating effects. However, the rapid diffusion and short retention time of H2 limit its direct clinical application. The development of H2-releasing biomaterials partially addresses these limitations and offers a synergistic approach that integrates gas therapy with bone tissue engineering. Although recent studies on H2-releasing systems for bone regeneration have markedly increased, a dedicated review in this field remains absent. To address this gap, this review summarizes the pathological features of bone defect microenvironments and the multi-target biological mechanisms of H2 relevant to bone repair. It then discusses design objectives, optimization strategies, and classifications of H2-releasing biomaterials. Recent advances in their applications for bone regeneration are highlighted with an emphasis on material design and functional performance. Cross-disciplinary insights are also introduced to inspire next-generation H2-releasing systems for bone regeneration. Finally, the limitations and clinical translation challenges of such systems are critically discussed, and future research directions are proposed. H2-releasing biomaterials integrate gas therapy, materials engineering, and microenvironmental regulation, representing an emerging frontier for multi-mechanism synergistic bone regeneration.
Accurate segmentation of gliomas in magnetic resonance imaging is crucial for clinical diagnosis and treatment planning. Existing segmentation methods based on three-dimensional MRI typically encounter dual challenges: insufficient sensitivity to tumor heterogeneity and inadequate characterization of regions with ambiguous boundaries. To address these challenges, we propose a novel dual-domain collaborative feature fusion network (DDCFF-Net). The proposed network adopts a three-branch architecture based on 3D U-Net: the high-frequency branch employs the non-subsampled contourlet transform, the low-frequency branch utilizes the dual-tree complex wavelet transform, and the spatial-domain branch directly processes the raw images. To further exploit frequency-domain information, an adaptive frequency convolution module and a frequency-domain cross-attention module are integrated to enhance feature representations at shallow and deep layers, respectively. Spatial-frequency fusion is accomplished via a dual-domain multi-scale fusion module and a progressive semantic-guided fusion module, refining semantic discriminability in a coarse-to-fine manner. Furthermore, a triple-supervision strategy combining supervised, unsupervised, and self-supervised learning is introduced to mitigate the distribution discrepancy between frequency-domain and spatial-domain features. Comprehensive evaluations on three public benchmarks validate that our method attains superior segmentation performance, with average Dice scores of 90.66%, 90.41%, and 89.88% on BraTS2021, BraTS2020, and BraTS2019, respectively. Notably, DDCFF-Net improves the Dice score by 1.10%(Avg) over the frequency-domain method HFF-Net and reduces the Hausdorff distance by 53.6% compared to Hyper-BTS on the enhancing tumor region. The proposed network effectively addresses tumor heterogeneity and boundary ambiguity challenges, making it well-suited for brain tumor diagnosis scenarios requiring high segmentation precision.
Immunotherapy serves as the fourth modality for cancer treatment, following surgery, radiotherapy, and chemotherapy. However, its efficacy remains inconsistent due to the heterogeneity of the tumor microenvironment (TME). Tumors characterized by an immunosuppressive TME that are unresponsive to immunotherapy are termed “cold tumors.” Thus, strategies to reverse this immunosuppressive environment are promising for treating cold tumors. Tumor-associated macrophages (TAMs) play a significant role in this process, making TAM-targeting therapies potentially valuable. Natural products (NPs), which are bioactive compounds from traditional herbs, exhibit broad anticancer effects and modulate TAM polarization, positioning them as a promising source for anticancer agents. This article explores the relationship between TAMs and the TME, emphasizing the mechanisms by which natural products, such as terpenoids and quinones, influence TAM polarization to inhibit tumor progression. By analyzing TAM roles in the TME and summarizing existing natural products targeting TAM polarization, this study offers insights for developing novel TAM-targeting drugs, and aids in the conversion of cold tumors to hot tumors.
