Lakshminarasimha Kasetty, Priyeshkumar A T, Karthikeyan M, Sakthivel Rajalaxmi
SVR Engineering College, Nandyal, India.
Biomedical Engineering, Mahendra College of Engineering, Salem, India.
Cancer Invest. 2025 Jun 23:1-20. doi: 10.1080/07357907.2025.2518404.
Lung cancer (LC) remains a leading cause of mortality worldwide, affecting individuals across all genders and age groups. Early and accurate diagnosis is critical for effective treatment and improved survival rates. Computed Tomography (CT) imaging is widely used for LC detection and classification. However, manual identification can be time-consuming and error-prone due to the visual similarities among various LC types. Deep learning (DL) has shown significant promise in medical image analysis. Although numerous studies have investigated LC detection using deep learning techniques, the effective extraction of highly correlated features remains a significant challenge, thereby limiting diagnostic accuracy. Furthermore, most existing models encounter substantial computational complexity and find it difficult to efficiently handle the high-dimensional nature of CT images. This study introduces an optimized CBAM-EfficientNet model to enhance feature extraction and improve LC classification. EfficientNet is utilized to reduce computational complexity, while the Convolutional Block Attention Module (CBAM) emphasizes essential spatial and channel features. Additionally, optimization algorithms including Gray Wolf Optimization (GWO), Whale Optimization (WO), and the Bat Algorithm (BA) are applied to fine-tune hyperparameters and boost predictive accuracy. The proposed model, integrated with different optimization strategies, is evaluated on two benchmark datasets. The GWO-based CBAM-EfficientNet achieves outstanding classification accuracies of 99.81% and 99.25% on the Lung-PET-CT-Dx and LIDC-IDRI datasets, respectively. Following GWO, the BA-based CBAM-EfficientNet achieves 99.44% and 98.75% accuracy on the same datasets. Comparative analysis highlights the superiority of the proposed model over existing approaches, demonstrating strong potential for reliable and automated LC diagnosis. Its lightweight architecture also supports real-time implementation, offering valuable assistance to radiologists in high-demand clinical environments.
肺癌(LC)仍然是全球主要的死亡原因,影响着所有性别和年龄组的人群。早期准确诊断对于有效治疗和提高生存率至关重要。计算机断层扫描(CT)成像广泛用于肺癌的检测和分类。然而,由于各种肺癌类型之间在视觉上存在相似性,人工识别可能既耗时又容易出错。深度学习(DL)在医学图像分析中已显示出巨大的潜力。尽管众多研究已使用深度学习技术来研究肺癌检测,但有效提取高度相关的特征仍然是一项重大挑战,从而限制了诊断准确性。此外,大多数现有模型都面临着巨大的计算复杂性,并且难以有效处理CT图像的高维特性。本研究引入了一种优化的CBAM-EfficientNet模型,以增强特征提取并改善肺癌分类。EfficientNet用于降低计算复杂性,而卷积块注意力模块(CBAM)则强调重要的空间和通道特征。此外,还应用了包括灰狼优化(GWO)、鲸鱼优化(WO)和蝙蝠算法(BA)在内的优化算法来微调超参数并提高预测准确性。所提出的模型与不同的优化策略相结合,在两个基准数据集上进行了评估。基于GWO的CBAM-EfficientNet在Lung-PET-CT-Dx和LIDC-IDRI数据集上分别实现了99.81%和99.25%的出色分类准确率。继GWO之后,基于BA的CBAM-EfficientNet在相同数据集上实现了99.44%和98.75%的准确率。对比分析突出了所提出模型相对于现有方法的优越性,显示出其在可靠且自动化的肺癌诊断方面具有强大潜力。其轻量级架构还支持实时实施,在高需求的临床环境中为放射科医生提供了宝贵的帮助。