Abstract Colorectal cancer (CRC) is a significant global health challenge, emphasizing the importance of effective screening by applying methods like colonoscopy. While advanced imaging technologies, such as narrow-band imaging (NBI), allow real-time optical diagnosis of colon polyps, variations in endoscopist skills and unnecessary polypectomy underscore the need for artificial intelligence applications, particularly deep learning (DL) in computer-aided polyp detection and diagnosis (CADe and CADx). This study developed and investigated a data augmentation technique using specular reflection (SR) to enhance the robustness and performance of DL models tailored explicitly for CADx in colonoscopy. This SR augmentation method included SR generation and inpainting integrated into conventional augmentation techniques. We utilized two DL architectures: a convolutional neural network and a vision transformer. Stress tests, under varying data usage ratios using a dataset of 2,616 NBI images, revealed the robustness of SR augmentation, especially in scenarios with limited training data, highlighting its superiority over other methods. SR augmentation effectively improved model accuracy, particularly in scenarios with limited data, supporting its practical implementation in real-world colonoscopy environments. The findings emphasize the significance of domain-specific data augmentation techniques to support DL application in colonoscopy imaging for more reliable and accurate CADx systems for colon polyps.

Cardiovascular disease stands as the leading cause of death globally, and substantial research has revealed its close correlation with the distribution of epicardial adipose tissue (EAT). Moreover, existing studies have demonstrated that EAT exhibits significant differences in distribution patterns and pathophysiological roles across various anatomical regions of the heart. Therefore, the quantitative analysis of EAT at different cardiac locations is crucial, and fine-grained segmentation of EAT via cardiac CT is an efficient method for obtaining the corresponding metrics. The existing computer-aided segmentation approaches typically treat EAT as a unified whole, which fails to meet the demands of nuanced diagnostics, and faces challenges such as class imbalance, thin structures, and anatomical variation, leading to low segmentation accuracy, limiting its application in cardiovascular disease risk assessment. To address these issues, we extend the existing segmentation strategy to the fine-grained segmentation of the left ventricle- (LV-), right ventricle- (RV-), and peri-atrium- (PA-) EAT, and propose the PRAEE framework based on position priors and edge enhancement. The core innovations of the proposed method are as follows: (1)Position-Prior Regularization: Considering the spatial distribution patterns of EAT in different anatomical regions, we design a regularization module that incorporates prior knowledge of typical spatial locations of various types of EAT as auxiliary constraints. This mechanism effectively guides the model to more accurately localize and differentiate EAT across anatomical regions, enabling an initial segmentation.(2)Adaptive Edge Enhancement: To further improve segmentation accuracy, we develop an edge enhancement module that explicitly extracts critical edge information through morphological operations and integrates it into the network architecture, significantly refining segmentation along boundary regions. Our approach has been validated on both a self-collected EAT dataset and the publicly available ACDC and MM-WHS datasets, demonstrating its effectiveness in enhancing fine-grained discrimination and edge detail performance.

Early detection of breast cancer significantly improves survival rates, with nearly all patients surviving for over five years. Mathematical modeling of cancerous tissue dynamics facilitates the rapid detection of tumors. This study introduces an innovative segmented hemispherical modeling approach for breast tissue, wherein the tissues are modeled as electrical capacitors with unequal plates. The structure and performance of the proposed hemispherical model are thoroughly examined. The effective permittivity, [Formula: see text], of both individual breast tissues and the entire breast is computed using their dielectric properties. The proposed closed-form breast model is analyzed and compared with state-of-the-art methods through analytical, simulation-based, and experimental approaches. The proposed segmented hemispherical modeling technique significantly outperforms traditional cubic models, achieving substantially higher discrimination levels of 0.335 compared to 0.001 for fatty breast tissue and 0.412 compared to 0.001 for dense breast tissue. The model accurately replicates real breast anatomy and demonstrates superior efficacy in tumor detection, showing a simulated difference of 3dB and 7 degrees in the magnitude and phase of the [Formula: see text]-parameters, respectively. Furthermore, the proposed method holds promising potential for developing an affordable and simple breast phantom fabrication method that, if adopted, could significantly facilitate research in laboratory settings. These phantoms would maintain high accuracy in replicating real breast tissue and contribute to more practical and reliable results in breast cancer detection techniques.

Schizophrenia (SCZ) is a severe and persistent mental health condition that profoundly affects individuals, their families, and broader communities. With rising global incidence and symptoms overlapping with disorders like bipolar illness, many remain unaware of its presence in daily life. Early diagnosis enables timely intervention, improving treatment outcomes and symptom management. Traditional machine learning approaches for schizophrenia detection rely on feature extraction and selection before classification. Deep learning (DL), renowned for modeling complex hierarchical patterns, accelerates the development of precise and objective diagnostic tools. Therefore, this research proposed a novel hybrid deep-learning approach for diagnosing Schizophrenia at an early stage. In this study, we developed an innovative framework employing the Mutation-enhanced Archimedes Optimization (MAO) algorithm to improve EEG preprocessing and signal clarity. Spatial and temporal features from multi-channel EEG data are analyzed through a hybrid deep learning approach, which mainly combines a Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) network. The proposed framework integrated an MAO into the CNN-GRU-MAO model, which enhances the capability to detect schizophrenia. A dual-objective optimization technique bootup detection accuracy and noise reduction, enhancing the overall effectiveness of the model. The experimental results demonstrated excellent performance and outperformed traditional approaches in terms of accuracy, precision, recall, F1-score, and specificity 98.41%, 98.13%, 98.87%, 98.49%, and 97.78% respectively. The MAO technique also evaluates signal integrity, enhancing Signal-to-Noise Ratio (SNR) and Signal-to-Interference Ratio (SIR) while reducing artifact contamination. This study highlights the ability of the MAO method in EEG preprocessing for schizophrenia detection. Integrating a deep learning framework with innovative optimization methods offers a transformative mechanism for improving mental health diagnostics via neurophysiological signal analysis.