Accurate colon polyp detection and segmentation in colonoscopy images remain a challenging task due to the variability in colon polyp appearance, size, shape, and location. In this paper, we propose EvoGrayNet, an automated evolutionary framework for colon polyp segmentation that facilitates detailed measurement of polyp boundaries, shape, size, and structural contrast. EvoGrayNet integrates a custom-designed feature extraction module with an evolutionary neural architecture search method to enhance spatial segmentation accuracy and model efficiency. First, we introduce a gray module that combines standard and depthwise separable convolutions, batch normalization, and dropout to capture both local details and global context in colonoscopy images. Second, a lightweight attention gate (AG) module is proposed to refine feature maps by focusing on low-contrast regions and improving localization accuracy. Third, we integrate the gray module and the AG module into a supernet along with a dilated feature extractor (DFE) and a feature recalibration (FR) module. The DFE captures multiscale spatial information using dilated convolutions, while the FR recalibrates channel importance for better feature representation. Finally, EvoGrayNet is optimized through an evolutionary process involving crossover and mutation, systematically identifying configurations that enhance boundary delineation accuracy. Performance is quantitatively assessed using the dice similarity coefficient (DSC), intersection over union (IoU), and the 95th percentile hausdorff distance (HD95), which together reflect both the precision of spatial measurement and boundary adherence. Experimental results show that EvoGrayNet outperforms 12 existing methods across four public colon polyp datasets, offering superior detection and segmentation accuracy, lower complexity, reduced FLOPS, faster inference speed, and improved generalization ability.

Diabetes remains a significant health challenge globally, contributing to severe complications like kidney disease, vision loss, and heart issues. The application of machine learning (ML) in healthcare enables efficient and accurate disease prediction, offering avenues for early intervention and patient support. Our study introduces an innovative diabetes prediction framework, leveraging both traditional ML techniques such as Logistic Regression, SVM, Naïve Bayes, and Random Forest and advanced ensemble methods like AdaBoost, Gradient Boosting, Extra Trees, and XGBoost. Central to our approach is the development of a novel model, DNet, a hybrid architecture combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) layers for effective feature extraction and sequential learning. The DNet model comprises an initial convolutional block for capturing essential features, followed by a residual block with skip connections to facilitate efficient information flow. Batch Normalization and Dropout are employed for robust regularization, and an LSTM layer captures temporal dependencies within the data. Using a Kaggle-sourced real-world diabetes dataset, our model evaluation spans cross-validation accuracy, precision, recall, F1 score, and ROC-AUC. Among the models, DNet demonstrates the highest efficacy with an accuracy of 99.79% and an AUC-ROC of 99.98%, establishing its potential for superior diabetes prediction. This robust hybrid architecture showcases the value of combining CNN and LSTM layers, emphasizing its applicability in medical diagnostics and disease prediction tasks.

Deep learning (DL) has reshaped the landscape of medical imaging by enabling automated, scalable, and high-accuracy analysis across diverse clinical modalities including MRI, CT, PET/SPECT, ultrasound, Xray, and digital pathology. Early convolutional neural networks (CNNs) established the foundation for tasks such as segmentation, detection, classification, and registration by learning hierarchical spatial representations directly from raw images. Subsequent architectural innovations particularly attention mechanisms, hybrid CNN-attention models, and transformer-based encoders have significantly improved the modeling of long-range dependencies, multi-scale anatomical structures, and complex clinical patterns. Neural architecture search (NAS) has further advanced the field by automating the design and optimization of medical AI systems, leading to architectures that are more efficient, adaptive, and robust. Parallel progress in self-supervised learning and generative modeling, including masked autoencoding, contrastive learning, GANs, VAEs, and diffusion models, has reduced the dependence on large annotated datasets while improving representation quality, modality translation, superresolution, and data augmentation. Multi-modal fusion techniques now enable the integration of imaging, clinical variables, and genomics, supporting more holistic and patient-specific decision making. However, significant challenges persist in clinical deployment, including domain shift, acquisition variability, data imbalance, uncertainty quantification, fairness, out-of-distribution detection, and regulatory requirements for safety and interpretability. This chapter presents a comprehensive overview of these advances, covering foundational principles, architectural evolution, generative and self-supervised paradigms, robustness and generalization strategies, and applications across neuroimaging, oncology, cardiology, abdominal and gastrointestinal imaging, ophthalmology, and pathology. Finally, emerging directions such as foundation models, federated learning, harmonization, and digital-twin frameworks are discussed as key enablers of next generation medical AI systems that are reliable, equitable, and clinically meaningful.