Magnetic Resonance Imaging (MRI) based three-dimensional analysis of knee cartilage has evolved to become fully automatic. However, when implementing these measurements across multiple clinical centers, scanner variability becomes a critical consideration. Our purposes were to quantify and compare same-model variability (between repeated scans on the same MRI system) and cross-model variability (across different MRI systems) in knee cartilage thickness measurements using MRI scanners from five manufacturers, as analyzed with a specific 3D volume analysis software. Ten healthy volunteers (eight males and two females, aged 22-60 years) underwent two scans of their right knee on 3T MRI systems from five manufacturers (Canon, Fujifilm, GE, Philips, and Siemens). The imaging protocol included fat-suppressed spoiled gradient echo and proton density weighted sequences. Cartilage regions were automatically segmented into 7 subregions using a specific deep learning-based 3D volume analysis software. This resulted in 350 measurements for same-model variability and 2,800 measurements for cross-model variability. For same-model variability, 82% of measurements showed variability ≤0.10 mm, and 98% showed variability ≤0.20 mm. For cross-model variability, 51% showed variability ≤0.10 mm, and 84% showed variability ≤0.20 mm. The mean same-model variability (0.06 ± 0.05 mm) was significantly lower than cross-model variability (0.11 ± 0.09 mm) (p < 0.001). This study demonstrates that knee cartilage thickness measurements exhibit significantly higher variability across different MRI systems compared to repeated measurements on the same system, when analyzed using this specific software. This finding has important implications for multi-center studies and longitudinal assessments using different MRI systems and highlights the software-dependent nature of such variability assessments.

Optimizing resource allocation for Parkinson's disease (PD) motor rehabilitation necessitates identifying biomarkers of responsiveness and dynamic neuroplasticity signatures underlying efficacy. A cohort study of 52 early-stage PD patients undergoing 2-week multidisciplinary intensive rehabilitation therapy (MIRT) was conducted, which stratified participants into responders and nonresponders. A multimodal analysis of resting-state electroencephalography (EEG) microstates and functional magnetic resonance imaging (fMRI) coactivation patterns was performed to characterize MIRT-induced spatiotemporal network reorganization. Responders demonstrated clinically meaningful improvement in motor symptoms, exceeding the minimal clinically important difference threshold of 3.25 on the Unified PD Rating Scale part III, alongside significant reductions in bradykinesia and a significant enhancement in quality-of-life scores at the 3-month follow-up. Resting-state EEG in responders showed a significant attenuation in microstate C and a significant enhancement in microstate D occurrences, along with significantly increased transitions from microstate A/B to D, which significantly correlated with motor function, especially in bradykinesia gains. Concurrently, fMRI analyses identified a prolonged dwell time of the dorsal attention network coactivation/ventral attention network deactivation pattern, which was significantly inversely associated with microstate C occurrence and significantly linked to motor improvement. The identified brain spatiotemporal neural markers were validated using machine learning models to assess the efficacy of MIRT in motor rehabilitation for PD patients, achieving an average accuracy rate of 86%. These findings suggest that MIRT may facilitate a shift in neural networks from sensory processing to higher-order cognitive control, with the dynamic reallocation of attentional resources. This preliminary study validates the necessity of integrating cognitive-motor strategies for the motor rehabilitation of PD and identifies novel neural markers for assessing treatment efficacy.

<b>Background:</b> No robust biomarkers have been identified to predict the efficacy of programmed cell death protein 1 (PD-1) inhibitors in patients with locoregionally advanced nasopharyngeal carcinoma (LANPC). We aimed to develop radiomic models using pre-immunotherapy MRI to predict the response to PD-1 inhibitors and the patient prognosis. <b>Methods:</b> This study included 246 LANPC patients (training cohort, <i>n</i> = 117; external test cohort, <i>n</i> = 129) from 10 centers. The best-performing machine learning classifier was employed to create the radiomic models. A combined model was constructed by integrating clinical and radiomic data. A radiomic interpretability study was performed with whole slide images (WSIs) stained with hematoxylin and eosin (H&E) and immunohistochemistry (IHC). A total of 150 patient-level nuclear morphological features (NMFs) and 12 cell spatial distribution features (CSDFs) were extracted from WSIs. The correlation between the radiomic and pathological features was assessed using Spearman correlation analysis. <b>Results:</b> The radiomic model outperformed the clinical and combined models in predicting treatment response (area under the curve: 0.760 vs. 0.559 vs. 0.652). For overall survival estimation, the combined model performed comparably to the radiomic model but outperformed the clinical model (concordance index: 0.858 vs. 0.812 vs. 0.664). Six treatment response-related radiomic features correlated with 50 H&E-derived (146 pairs, |<i>r</i>|= 0.31 to 0.46) and 2 to 26 IHC-derived NMF, particularly for CD45RO (69 pairs, |<i>r</i>|= 0.31 to 0.48), CD8 (84, |<i>r</i>|= 0.30 to 0.59), PD-L1 (73, |<i>r</i>|= 0.32 to 0.48), and CD163 (53, |<i>r</i>| = 0.32 to 0.59). Eight prognostic radiomic features correlated with 11 H&E-derived (16 pairs, |<i>r</i>|= 0.48 to 0.61) and 2 to 31 IHC-derived NMF, particularly for PD-L1 (80 pairs, |<i>r</i>|= 0.44 to 0.64), CD45RO (65, |<i>r</i>|= 0.42 to 0.67), CD19 (35, |<i>r</i>|= 0.44 to 0.58), CD66b (61, |<i>r</i>| = 0.42 to 0.67), and FOXP3 (21, |<i>r</i>| = 0.41 to 0.71). In contrast, fewer CSDFs exhibited correlations with specific radiomic features. <b>Conclusion:</b> The radiomic model and combined model are feasible in predicting immunotherapy response and outcomes in LANPC patients. The radiology-pathology correlation suggests a potential biological basis for the predictive models.

Approaches studying the dynamics of resting-state functional magnetic resonance imaging (rs-fMRI) activity often focus on time-resolved functional connectivity (tr-FC). While many tr-FC approaches have been proposed, most are linear approaches, e.g. computing the linear correlation at a timestep or within a window. In this work, we propose to use a generative non-linear deep learning model, a disentangled variational autoencoder (DSVAE), that factorizes out window-specific (context) information from timestep-specific (local) information. This has the advantage of allowing our model to capture differences at multiple temporal scales. We find that by separating out temporal scales our model's window-specific embeddings, or as we refer to them, context embeddings, more accurately separate windows from schizophrenia patients and control subjects than baseline models and the standard tr-FC approach in a low-dimensional space. Moreover, we find that for individuals with schizophrenia, our model's context embedding space is significantly correlated with both age and symptom severity. Interestingly, patients appear to spend more time in three clusters, one closer to controls which shows increased visual-sensorimotor, cerebellar-subcortical, and reduced cerebellar-visual functional network connectivity (FNC), an intermediate station showing increased subcortical-sensorimotor FNC, and one that shows decreased visual-sensorimotor, decreased subcortical-sensorimotor, and increased visual-subcortical domains. We verify that our model captures features that are complementary to - but not the same as - standard tr-FC features. Our model can thus help broaden the neuroimaging toolset in analyzing fMRI dynamics and shows potential as an approach for finding psychiatric links that are more sensitive to individual and group characteristics.

Currently, there are inconsistencies among different studies on preoperative prediction of Cytokeratin 19 (CK19) expression in HCC using traditional imaging, radiomics, and deep learning. We aimed to systematically analyze and compare the performance of non-invasive methods for predicting CK19-positive HCC, thereby providing insights for the stratified management of HCC patients. A comprehensive literature search was conducted in PubMed, EMBASE, Web of Science, and the Cochrane Library from inception to February 2025. Two investigators independently screened and extracted data based on inclusion and exclusion criteria. Eligible studies were included, and key findings were summarized in tables to provide a clear overview. Ultimately, 22 studies involving 3395 HCC patients were included. 72.7% (16/22) focused on traditional imaging, 36.4% (8/22) on radiomics, 9.1% (2/22) on deep learning, and 54.5% (12/22) on combined models. The magnetic resonance imaging was the most commonly used imaging modality (19/22), and over half of the studies (12/22) were published between 2022 and 2025. Moreover, 27.3% (6/22) were multicenter studies, 36.4% (8/22) included a validation set, and only 13.6% (3/22) were prospective. The area under the curve (AUC) range of using clinical and traditional imaging was 0.560 to 0.917. The AUC ranges of radiomics were 0.648 to 0.951, and the AUC ranges of deep learning were 0.718 to 0.820. Notably, the AUC ranges of combined models of clinical, imaging, radiomics and deep learning were 0.614 to 0.995. Nevertheless, the multicenter external data were limited, with only 13.6% (3/22) incorporating validation. The combined model integrating traditional imaging, radiomics and deep learning achieves excellent potential and performance for predicting CK19 in HCC. Based on current limitations, future research should focus on building an easy-to-use dynamic online tool, combining multicenter-multimodal imaging and advanced deep learning approaches to enhance the accuracy and robustness of model predictions.

Non-contrast Computed Tomography (NCCT) quickly diagnoses acute cerebral hemorrhage or infarction. However, Deep-Learning (DL) algorithms often generate false alarms (FA) beyond the cerebral region. We introduce an enhanced brain tissue segmentation method for infarction lesion segmentation (ILS). This method integrates an adaptive result fusion strategy to confine the search operation within cerebral tissue, effectively reducing FAs. By leveraging fused brain masks, DL-based ILS algorithms focus on pertinent radiomic correlations. Various U-Net models underwent rigorous training, with exploration of diverse fusion strategies. Further refinement entailed applying a 9x9 Gaussian filter with unit standard deviation followed by binarization to mitigate false positives. Performance evaluation utilized Intersection over Union (IoU) and Hausdorff Distance (HD) metrics, complemented by external validation on a subset of the COCO dataset. Our study comprised 20 ischemic stroke patients (14 males, 4 females) with an average age of 68.9 ± 11.7 years. Fusion with UNet2+ and UNet3 + yielded an IoU of 0.955 and an HD of 1.33, while fusion with U-net, UNet2 + , and UNet3 + resulted in an IoU of 0.952 and an HD of 1.61. Evaluation on the COCO dataset demonstrated an IoU of 0.463 and an HD of 584.1 for fusion with UNet2+ and UNet3 + , and an IoU of 0.453 and an HD of 728.0 for fusion with U-net, UNet2 + , and UNet3 + . Our adaptive fusion strategy significantly diminishes FAs and enhances the training efficacy of DL-based ILS algorithms, surpassing individual U-Net models. This methodology holds promise as a versatile, data-independent approach for cerebral lesion segmentation.

Medical image segmentation plays an important role in medical diagnosis and treatment. Most recent medical image segmentation methods are based on a convolutional neural network (CNN) or Transformer model. However, CNN-based methods are limited by locality, whereas Transformer-based methods are constrained by the quadratic complexity of attention computations. Alternatively, the state-space model-based Mamba architecture has garnered widespread attention owing to its linear computational complexity for global modeling. However, Mamba and its variants are still limited in their ability to extract local receptive field features. To address this limitation, we propose a novel residual spatial state-space (RSSS) block that enhances spatial feature extraction by integrating global and local representations. The RSSS block combines the Mamba module for capturing global dependencies with a receptive field attention convolution (RFAC) module to extract location-sensitive local patterns. Furthermore, we introduce a residual adjust strategy to dynamically fuse global and local information, improving spatial expressiveness. Based on the RSSS block, we design a U-shaped SA-UMamba segmentation framework that effectively captures multi-scale spatial context across different stages. Experiments conducted on the Synapse, ISIC17, ISIC18 and CVC-ClinicDB datasets validate the segmentation performance of our proposed SA-UMamba framework.

Acute ischemic stroke (AIS) has a poor prognosis and a high recurrence rate. Predicting the outcomes of AIS patients in the early stages of the disease is therefore important. The establishment of intracerebral collateral circulation significantly improves the survival of brain cells and the outcomes of AIS patients. However, no machine learning method has been applied to investigate the correlation between the dynamic evolution of intracerebral venous collateral circulation and AIS prognosis. Therefore, we employed a support vector machine (SVM) algorithm to analyze asymmetrically prominent cortical veins (APCVs) and deep medullary veins (DMVs) to establish a radiomic model for predicting the prognosis of AIS by combining clinical indicators. The magnetic resonance imaging (MRI) data and clinical indicators of 150 AIS patients were retrospectively analyzed. Regions of interest corresponding to the DMVs and APCVs were delineated, and least absolute shrinkage and selection operator (LASSO) regression was used to select features extracted from these regions. An APCV-DMV radiomic model was created via the SVM algorithm, and independent clinical risk factors associated with AIS were combined with the radiomic model to generate a joint model. The SVM algorithm was selected because of its proven efficacy in handling high-dimensional radiomic data compared with alternative classifiers (<i>e.g.</i>, random forest) in pilot experiments. Nine radiomic features associated with AIS patient outcomes were ultimately selected. In the internal training test set, the AUCs of the clinical, DMV-APCV radiomic and joint models were 0.816, 0.976 and 0.996, respectively. The DeLong test revealed that the predictive performance of the joint model was better than that of the individual models, with a test set AUC of 0.996, sensitivity of 0.905, and specificity of 1.000 (<i>P</i> < 0.05). Using radiomic methods, we propose a novel joint predictive model that combines the imaging histologic features of the APCV and DMV with clinical indicators. This model quantitatively characterizes the morphological and functional attributes of venous collateral circulation, elucidating its important role in accurately evaluating the prognosis of patients with AIS and providing a noninvasive and highly accurate imaging tool for early prognostic prediction.

Despite significant progress in applying large language models (LLMs) to the medical domain, several limitations still prevent them from practical applications. Among these are the constraints on model size and the lack of cohort-specific labeled datasets. In this work, we investigated the potential of improving a lightweight LLM, such as Llama 3.1-8B, through fine-tuning with datasets using synthetic labels. Two tasks are jointly trained by combining their respective instruction datasets. When the quality of the task-specific synthetic labels is relatively high (e.g., generated by GPT4-o), Llama 3.1-8B achieves satisfactory performance on the open-ended disease detection task, with a micro F1 score of 0.91. Conversely, when the quality of the task-relevant synthetic labels is relatively low (e.g., from the MIMIC-CXR dataset), fine-tuned Llama 3.1-8B is able to surpass its noisy teacher labels (micro F1 score of 0.67 v.s. 0.63) when calibrated against curated labels, indicating the strong inherent underlying capability of the model. These findings demonstrate the potential offine-tuning LLMs with synthetic labels, offering a promising direction for future research on LLM specialization in the medical domain.

Lung cancer is a major cause of cancer-related deaths, and early diagnosis and treatment are crucial for improving patients' survival outcomes. In this paper, we propose to employ convolutional neural networks to model the non-linear relationship between the risk of lung cancer and the lungs' morphology revealed in the CT images. We apply a mini-batched loss that extends the Cox proportional hazards model to handle the non-convexity induced by neural networks, which also enables the training of large data sets. Additionally, we propose to combine mini-batched loss and binary cross-entropy to predict both lung cancer occurrence and the risk of mortality. Simulation results demonstrate the effectiveness of both the mini-batched loss with and without the censoring mechanism, as well as its combination with binary cross-entropy. We evaluate our approach on the National Lung Screening Trial data set with several 3D convolutional neural network architectures, achieving high AUC and C-index scores for lung cancer classification and survival prediction. These results, obtained from simulations and real data experiments, highlight the potential of our approach to improving the diagnosis and treatment of lung cancer.