<p>Lung cancer presents a substantial worldwide health concern, standing as one of the leading causes of cancer-related fatalities. Computed tomography (CT) imaging has become vital for diagnosing lung diseases, especially in identifying pulmonary nodules early on, which is important for planning treatment. Detecting small nodules with variations in CT images and monitoring them promptly raises challenges due to the intricate structure of medical data, which could result in misclassification between benign and malignant nodules. This research addresses the impact of these challenges by proposing an enhanced one-stage detector and a two-stage detector for the segmentation of pulmonary nodules. By implementing a modified YOLOv8, a one-stage detector, incorporating a squeeze and excitation network in the head structures to enhance feature channels by recalibrating the feature map utilizing channel-wise multiplication. Furthermore, a modified mask R-CNN, a two-stage detector, was proposed by incorporating the channel block attention mechanism (CBAM) into the region proposal network to enhance feature extraction during the semantic extraction phase. The dataset used in this study comprises the IQ-OTH/NCCD benchmark dataset and clinical data obtained from the&#xa0;SRM Medical Hospital. The experimental results show that the improved two-stage detector attained a precision that increased from 89% to 92.6%, recall from 87.2% to 90.1%, and mAP from 88.9% to 91.5%. The modified one-stage detector demonstrated a faster detection time, with 34&#xa0;ms for benign nodules and 20&#xa0;ms for malignant lesions. The attention mechanism enhanced the region proposal and refining phases, yielding accurate detection of small lesions in CT images.</p>