Attention Enhanced InceptionNeXt-Based Hybrid Deep Learning Model for Lung Cancer Detection

Abstract

Lung cancer is the leading cause of cancer death which demands timely and accurate diagnosis to improve patient outcomes. Computed tomography (CT) scans are often poorly localized and identified using manual interpretation and standard deep learning methods to find multi-scale features. Two publicly available datasets, which are the IQ-OTH/NCCD Lung Cancer Dataset and the Chest CT-Scan Images Dataset are used to perform experiments. It is proposed to use the hybrid attention-enhanced CNNViT framework based on ResNet50, DenseNet169, EfficientNetV2-Medium, ConvNeXt-Base, InceptionNeXt-Base, MobileViT-Small, ConViT-Base, Swin-Base, MaxViT-Base, and DeiT3-Base to classify images, and YOLOv5, YO Preprocessing includes image resizing to 299x299, image data augmentation, normalizing tensors and stratified data division, at the same time YOLO datasets are cleaned with bounding box labels. GradCAM generates heatmaps highlighting noteworthy regions, whereas a Flask-based interface allows a thorough interaction with the user. ConvNeXt-Base has a peak classification rate of 99.09% on the IQ-OTH/NCCD data, and InceptionNeXt-Base has a peak accuracy rate of 99.01% on the chest CT-scan dataset. The highest mean average precision (mAP) of 72.8% is achieved with YOLOv5 in detection. The approach exhibits enhanced robustness, equitable performance, and elucidated predictions by the integration of categorization and detection inside a cohesive system.