The adoption of Transformer-based architectures in the medical domain is growing rapidly. In medical imaging, the analysis of complex shapes - such as organs, tissues, or other anatomical structures - combined with the often anisotropic nature of high-dimensional images complicates these adaptations. In this study, we critically examine the role of Positional Encodings (PEs), arguing that commonly used approaches may be suboptimal for the specific challenges of medical imaging. Sinusoidal Positional Encodings (SPEs) have proven effective in vision tasks, but they struggle to preserve Euclidean distances in higher-dimensional spaces. Isotropic Fourier Feature Positional Encodings (IFPEs) have been proposed to better preserve Euclidean distances, but they lack the ability to account for anisotropy in images. To address these limitations, we propose Anisotropic Fourier Feature Positional Encoding (AFPE), a generalization of IFPE that incorporates anisotropic, class-specific, and domain-specific spatial dependencies. We systematically benchmark AFPE against commonly used PEs on multi-label classification in chest X-rays, organ classification in CT images, and ejection fraction regression in echocardiography. Our results demonstrate that choosing the correct PE can significantly improve model performance. We show that the optimal PE depends on the shape of the structure of interest and the anisotropy of the data. Finally, our proposed AFPE significantly outperforms state-of-the-art PEs in all tested anisotropic settings. We conclude that, in anisotropic medical images and videos, it is of paramount importance to choose an anisotropic PE that fits the data and the shape of interest.

Soft  tissue sarcomas (STS) are heterogeneous malignancies with high recurrence rates (33-39%) post-surgery, necessitating improved prognostic tools. This study proposes a fusion model integrating deep transfer learning and radiomics from MRI to predict postoperative STS recurrence. Axial T2-weighted fat-suppressed imaging (T₂WI) of 803 STS patients from two institutions was retrospectively collected and divided into training (n = 527), internal validation (n = 132), and external validation (n = 144) cohorts. Tumor segmentation was performed using the SegResNet model within the Auto3DSeg framework. Radiomic features and deep learning features were extracted. Feature selection employed LASSO regression, and the deep learning radiomic (DLR) model combined radiomic and deep learning signatures. Using the features, nine models were constructed based on three classifiers. The area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, negative predictive value, and positive predictive value were calculated for performance evaluation. The SegResNet model achieved Dice coefficients of 0.728 after refinement. Recurrence rates were 22.8% (120/527) in the training, 25.0% (33/132) in the internal validation, and 32.6% (47/144) in the external validation cohorts. The DLR model (ExtraTrees) demonstrated superior performance, achieving an AUC of 0.818 in internal validation and 0.809 in external validation, better than the radiomic model (0.710, 0.612) and the deep learning model (0.751, 0.667). Sensitivity and specificity ranged from 0.702 to 0.976 and 0.732 to 0.830, respectively. Decision curve analysis confirmed superior clinical utility. The DLR model provides a robust, non-invasive tool for preoperative STS recurrence prediction, enabling personalized treatment decisions and postoperative management.

Despite the widespread success of convolutional neural networks (CNNs) in general computer vision tasks, their application to complex medical image analysis faces persistent challenges. These include limited labeled data availability, which restricts model generalization; class imbalance, where minority classes are underrepresented and lead to biased predictions; and inadequate feature representation, since conventional CNNs often struggle to capture subtle patterns and intricate dependencies characteristic of medical imaging. To address these limitations, we propose CTNGAN, a unified framework that integrates generative modeling with Generative Adversarial Networks (GANs), contrastive learning, and Transformer architectures to enhance the robustness and accuracy of medical image analysis. Each component is designed to tackle a specific challenge: the GAN model mitigates data scarcity and imbalance, contrastive learning strengthens feature robustness against domain shifts, and the Transformer captures long-range spatial patterns. This tripartite integration not only overcomes the limitations of conventional CNNs but also achieves superior generalizability, as demonstrated by classification experiments on benchmark medical imaging datasets, with up to 98.5% accuracy and an F1-score of 0.968, outperforming existing methods. The framework's ability to jointly optimize data generation, feature discrimination, and contextual modeling establishes a new paradigm for accurate and reliable medical image diagnosis.

Light-chain cardiac amyloidosis (AL-CA) is a progressive heart disease with high mortality rate and variable prognosis. Presently used Mayo staging method can only stratify patients into four stages, highlighting the necessity for a more individualized prognosis prediction method. We aim to develop a novel deep learning (DL) model for whole-heart analysis of cardiovascular magnetic resonance-derived late gadolinium enhancement (LGE) images to predict individualized prognosis in AL-CA. This study included 394 patients with AL-CA who underwent standardized chemotherapy and had at least one year of follow-up. The approach involved automated segmentation of heart in LGE images and feature extraction using a Transformer-based DL model. To enhance feature differentiation and mitigate overfitting, a contrastive pretraining strategy was employed to accentuate distinct features between patients with different prognosis while clustering similar cases. Finally, an ensemble learning strategy was used to integrate predictions from 15 models at 15 survival time points into a comprehensive prognostic model. In the testing set of 79 patients, the DL model achieved a C-Index of 0.91 and an AUC of 0.95 in predicting 2.6-year survival (HR: 2.67), outperforming the Mayo model (C-Index=0.65, AUC=0.71). The DL model effectively distinguished patients with the same Mayo stage but different prognosis. Visualization techniques revealed that the model captures complex, high-dimensional prognostic features across multiple cardiac regions, extending beyond the amyloid-affected areas. This fully automated DL model can predict individualized prognosis of AL-CA through LGE images, which complements the presently used Mayo staging method.

Autism Spectrum Disorder (ASD) is a neurological condition that affects the brain, leading to challenges in speech, communication, social interaction, repetitive behaviors, and motor skills. This research aims to develop a deep learning based model for the accurate diagnosis and classification of autistic symptoms in children, thereby benefiting both patients and their families. Existing literature indicates that classification methods typically analyze region based summaries of Functional Magnetic Resonance Imaging (fMRI). However, few studies have explored the diagnosis of ASD using brain imaging. The complexity and heterogeneity of biomedical data modeling for big data analysis related to ASD remain unclear. In the present study, the Autism Brain Imaging Data Exchange 1 (ABIDE-1) dataset was utilized, comprising 1,112 participants, including 539 individuals with ASD and 573 controls from 17 different sites. The dataset, originally in NIfTI format, required conversion to a computer-readable extension. For ASD classification, the researcher proposed and implemented a VGG20 architecture. This deep learning VGG20 model was applied to neuroimages to distinguish ASD from the non ASD cases. Four evaluation metrics were employed which are recall, precision, F1-score, and accuracy. Experimental results indicated that the proposed model achieved an accuracy of 61%. Prior to this work, machine learning algorithms had been applied to the ABIDE-1 dataset, but deep learning techniques had not been extensively utilized for this dataset and the methods implied in this study as this research is conducted to facilitate the early diagnosis of ASD.

Accurate breast MRI lesion detection is critical for early cancer diagnosis, especially in high-risk populations. We present a classification pipeline that adapts a pretrained foundation model, the Medical Slice Transformer (MST), for breast lesion classification using dynamic contrast-enhanced MRI (DCE-MRI). Leveraging DINOv2-based self-supervised pretraining, MST generates robust per-slice feature embeddings, which are then used to train a Kolmogorov--Arnold Network (KAN) classifier. The KAN provides a flexible and interpretable alternative to conventional convolutional networks by enabling localized nonlinear transformations via adaptive B-spline activations. This enhances the model's ability to differentiate benign from malignant lesions in imbalanced and heterogeneous clinical datasets. Experimental results demonstrate that the MST+KAN pipeline outperforms the baseline MST classifier, achieving AUC = 0.80 \pm 0.02 while preserving interpretability through attention-based heatmaps. Our findings highlight the effectiveness of combining foundation model embeddings with advanced classification strategies for building robust and generalizable breast MRI analysis tools.

In this paper, we propose a novel generative model to produce high-quality SAH samples, enhancing SAH CT detection performance in imbalanced datasets. Previous methods, such as cost-sensitive learning and previous diffusion models, suffer from overfitting or noise-induced distortion, limiting their effectiveness. Accurate SAH sample generation is crucial for better detection. We propose the Worley-Perlin Diffusion Model (WPDM), leveraging Worley-Perlin noise to synthesize diverse, high-quality SAH images. WPDM addresses limitations of Gaussian noise (homogeneity) and Simplex noise (distortion), enhancing robustness for generating SAH images. Additionally, <math xmlns="http://www.w3.org/1998/Math/MathML"><msub><mtext>WPDM</mtext> <mtext>Fast</mtext></msub> </math> optimizes generation speed without compromising quality. WPDM effectively improved classification accuracy in datasets with varying imbalance ratios. Notably, a classifier trained with WPDM-generated samples achieved an F1-score of 0.857 on a 1:36 imbalance ratio, surpassing the state of the art by 2.3 percentage points. WPDM overcomes the limitations of Gaussian and Simplex noise-based models, generating high-quality, realistic SAH images. It significantly enhances classification performance in imbalanced settings, providing a robust solution for SAH CT detection.

Deep learning has become a powerful tool for medical image analysis; however, conventional Convolutional Neural Networks (CNNs) often fail to capture the fine-grained and complex features critical for accurate diagnosis. To address this limitation, we systematically integrate attention mechanisms into five widely adopted CNN architectures, namely, VGG16, ResNet18, InceptionV3, DenseNet121, and EfficientNetB5, to enhance their ability to focus on salient regions and improve discriminative performance. Specifically, each baseline model is augmented with either a Squeeze and Excitation block or a hybrid Convolutional Block Attention Module, allowing adaptive recalibration of channel and spatial feature representations. The proposed models are evaluated on two distinct medical imaging datasets, a brain tumor MRI dataset comprising multiple tumor subtypes, and a Products of Conception histopathological dataset containing four tissue categories. Experimental results demonstrate that attention augmented CNNs consistently outperform baseline architectures across all metrics. In particular, EfficientNetB5 with hybrid attention achieves the highest overall performance, delivering substantial gains on both datasets. Beyond improved classification accuracy, attention mechanisms enhance feature localization, leading to better generalization across heterogeneous imaging modalities. This work contributes a systematic comparative framework for embedding attention modules in diverse CNN architectures and rigorously assesses their impact across multiple medical imaging tasks. The findings provide practical insights for the development of robust, interpretable, and clinically applicable deep learning based decision support systems.

Alzheimer's disease (AD) is a leading cause of dementia worldwide, characterized by heterogeneous neuropathological changes and progressive cognitive decline. Despite the numerous studies, there are still no effective treatments beyond those that aim to slow progression and compensate the impairment. Neuroimaging techniques provide a comprehensive view of brain changes, with magnetic resonance imaging (MRI) playing a key role due to its non-invasive nature and wide availability. The T1-weighted MRI sequence is frequently used due to its prevalence in most MRI protocols, generating large datasets, ideal for artificial intelligence (AI) applications. AI, particularly machine learning (ML) and deep learning (DL) techniques, has been increasingly utilized to model these datasets and classify individuals along the AD continuum. This systematic review evaluates studies using AI to classify more than two stages of AD based on T1-weighted MRI data. Convolutional neural networks (CNNs) are the most widely applied, achieving an average classification accuracy of 85.93 % (range: 51.80-100 %; median: 87.70 %). These good results are due to CNNs' ability to extract hierarchical features directly from raw imaging data, reducing the need for extensive preprocessing. Non-convolutional neural networks and traditional ML approaches also demonstrated strong performance, with mean accuracies of 82.50 % (range: 57.61-99.38 %; median: 86.67 %) and 84.22 % (range: 33-99.10 %; median: 87.75 %), respectively, underscoring importance of input data selection. Despite promising outcomes, challenges remain, including methodological heterogeneity, overfitting risks, and a reliance on the ADNI database, which limits dataset diversity. Addressing these limitations is critical to advancing AI's clinical application for early detection, improved classification, and enhanced patient outcomes.

Locally advanced rectal cancer (LARC) exhibits significant heterogeneity in response to neoadjuvant chemotherapy (NAC), with poor responders facing delayed treatment and unnecessary toxicity. Although MRI provides spatial pathophysiological information and proteomics reveals molecular mechanisms, current single-modal approaches cannot integrate these complementary perspectives, resulting in limited predictive accuracy and biological insight. This retrospective study developed a multimodal deep learning framework using a cohort of 274 LARC patients treated with NAC (2012-2021). Graph neural networks analyzed proteomic profiles from FFPE tissues, incorporating KEGG/GO pathways and PPI networks, while a spatially enhanced 3D ResNet152 processed T2WI. A LightGBM classifier integrated both modalities with clinical features using zero-imputation for missing data. Model performance was assessed through AUC-ROC, decision curve analysis, and interpretability techniques (SHAP and Grad-CAM). The integrated model achieved superior NAC response prediction (test AUC 0.828, sensitivity 0.875, specificity 0.750), significantly outperforming single-modal approaches (MRI ΔAUC +0.109; proteomics ΔAUC +0.125). SHAP analysis revealed MRI-derived features contributed 57.7% of predictive power, primarily through peritumoral stromal heterogeneity quantification. Proteomics identified 10 key chemoresistance proteins, including CYBA, GUSB, ATP6AP2, DYNC1I2, DAD1, ACOX1, COPG1, FBP1, DHRS7, and SSR3. Decision curve analysis confirmed clinical utility across threshold probabilities (0-0.75). Our study established a novel MRI-proteomics integration framework for NAC response prediction, with MRI defining spatial resistance patterns and proteomics deciphering molecular drivers, enabling early organ preservation strategies. The zero-imputation design ensured deplorability in diverse clinical settings.