Deep learning-based classification of colonoscopic images using an attention-enhanced ConvNeXt V2 architecture

An Explainable CBAM Enhanced DenseNet121 Framework for Multi-Class Lung Cancer Classification Using CT Scans

Due to its late identification and challenging diagnosis, lung cancer continues to be one of the top causes of death for cancer patients globally, positioning it as one of the most critical concerns. Timely identification of cancerous nodules is essential for enhancing the patient’s survival likelihood CT image analysis by hand is not very productive and significantly relies on a specialist’s expertise. In this study, we offer an autonomous lung cancer classification method based on explainable deep learning. The popular DenseNet121 network serves as the foundation for our deep learning model, which is enhanced by the Convolutional Block Attention Module (CBAM). To improve feature extraction of significant spatial and channel properties of input data, attention techniques are added. Furthermore, our method is interpretable because the Grad-CAM technique makes it possible to explain the choices made by a machine learning system. A database of CT scans, comprising 4,598 pictures categorized by large cell carcinoma, adenocarcinoma, and healthy lungs, was utilized. Our evaluations show the model’s effectiveness with an accuracy rate of 94.6\%.

Attention-Enhanced Deep Learning Frameworks for Accurate Lung Cancer Detection and Classification

Lung cancer is a life-threatening illness and early and proper diagnosis is essential to successful treatment and higher survival rates. Conventional deep learning systems, such as standard YOLO architectures, do not tend to differentiate between visually similar categories like benign and malignant nodules because of constraints in the representation of multi-scale features and understanding of context. To solve these two new modules are proposed in this work, the C3K2 and A2C2F modules in the YOLOv12x framework. C3K2 module enhances multi-kernel convolutional learning with both fine-grained and global spatial features that enhance the ability of the model to distinguish subtle structural differences in lung nodules. At the same time, A2C2F module (Adaptive Attention Cross-Channel Fusion) adapts the channel-wise features representations and improves the information flow between the layers. The experimental assessment of a benchmark lung CT images dataset shows that the proposed YOLOv12x model has better performance compared to the current YOLOv8x architecture. The C3K2 and A2C2F modules are important to the point that they increase feature discrimination, robustness and detection accuracy without decreasing computational efficiency.

A Multi-head Attention Transformer with Convolutional Projections for Skin Cancer Image Classification

ATTENTION-BASED NETWORK FOR COLORECTAL CANCER CLASSIFICATION USING MOBILENETV2

Colorectal cancer (CRC) is among the leading causes of cancer-related deaths worldwide, emphasizing the importance of accurate and timely diagnosis. Histopathological image analysis remains the gold standard for colorectal cancer diagnosis; however, manual examination is labor-intensive and prone to inter-observer variability. Recent advances in deep learning have significantly improved automated cancer classification by leveraging convolutional neural networks (CNNs). Nevertheless, conventional CNN architectures often fail to capture the most discriminative regions of tissue images, resulting in suboptimal classification performance. This paper proposes an Attention-Based MobileNetV2 (AB-MobileNetV2) framework that integrates a lightweight MobileNetV2 backbone with an attention mechanism to enhance feature representation for colorectal cancer classification. The proposed architecture employs channel and spatial attention modules to emphasize informative pathological regions while suppressing irrelevant background information. Extensive experiments conducted on benchmark colorectal histopathological datasets demonstrate that the proposed model achieves superior classification performance compared to conventional CNN models while maintaining computational efficiency suitable for real-time clinical applications. Experimental results show an overall accuracy of 98.92%, precision of 98.75%, recall of 98.63%, F1-score of 98.69%, and AUC of 99.34%, outperforming several state-of-the-art approaches.

Skin Disease Detection using a Hybrid CNN and Vision Transformer Model

Skin disease classification through deep learning has emerged as an important research field because of its ability to facilitate the accurate and timely medical diagnosis. A hybrid deep learning model combining Convolutional Neural Networks (CNN) and a Vision Transformer (ViT) is introduced in the present study to classify skin diseases automatically from dermoscopic images. Images selected are dermoscopic for their ability to provide detailed information about the boundary, distribution of pigmentation, and texture characteristics of lesions, key elements for accurate diagnosis. The experimental dataset is made up of 2000 dermoscopic images divided into 1400 images for model training and 600 images for testing purposes. The proposed framework is compared with the standalone CNN model, with the same experimental settings. Balanced sampling is combined with data augmentation techniques in order to reduce the impact of class imbalance within different categories of skin diseases during the training procedure. Common metrics used to evaluate the effectiveness of models are accuracy, precision, recall and F1-score. The experimental results have shown that the proposed Hybrid CNN–ViT model achieves 91.7% accuracy, which is higher than 84.1% accuracy obtained by the standalone CNN model. CNN-based local feature extraction is combined with the global contextual representation ability of the Vision Transformer, achieving the improved performance is due to the feature fusion between the two, which enables the model to have a better discrimination of visually similar skin lesions. The integration constitutes the main novelty of the proposed work. Moreover, the model is tested with the full test data to obtain the constant and reliable evaluation of the model. The hybrid architecture consumes more computational resources than a conventional CNN model, but the results show that it can achieve better accuracy in skin disease classification, strength, and reliability when detecting skin diseases automatically.

A Context-Adaptive Dual Cross-Attention Fusion Framework with Incremental Learning for Robust Lung and Colon Histopathology Classification

The accurate recognition of lung and colon histopathological images is still one of the most difficult tasks within computational pathology, mainly due to the morphological similarities of different malignancies. This study proposes a novel Context Adaptive Dual Cross Attention Feature Fusion with Incremental Learning framework for efficient multi-class histopathology image classification on the LC25000 dataset. Specifically, the authors' model combines deep spatial features from the VGG13 architecture with multiple handcrafted domain features, such as colour histograms, Local Binary Patterns (LBPs), and Grey-Level Co-Occurrence Matrices (GLCMs), using a token-based cross-attention module. To stabilise attention learning by considering the impact of tissue morphology on affinities, the authors propose a context-adaptive bias modulation strategy. Moreover, a token importance reweighting method is considered to increase the contribution of discriminative features. To address the problem of catastrophic forgetting in multi-class histopathological classification, the incremental learning method is applied. Experimentally, the authors' method achieved an accuracy of 0.9869, an F1-macro score of 0.9862, an AUC-macro score of 0.9924, and a Cohen's kappa of 0.9838. The proposed solution outperformed all baseline feature fusion approaches considered. In particular, the model demonstrated promising results for the lung subtype classification task.