Computed tomography (CT) scans are widely used for diagnosing lung infections, but manual interpretation is laborious. Artificial intelligence has spurred the development of efficient computer-aided diagnostic (CAD) systems, promising faster and more accurate diagnosis. However, many existing CAD systems lack sufficient cross-data analysis and consequently show suboptimal performance. To address their limitations, we propose a lightweight Meta-Domain Adoptive Segmentation Network (MDA-SN) with adaptive data normalization to enhance infection detection in cross-data analysis. Our optimal network design leverages multi-scale dilated grouped convolution with residual attention to ensure real-time performance and maintain accuracy. We further utilize the model to build a semantic attention-driven retrieval framework, enabling infection ratio quantification and retrieval of relevant CT slices from the database, closely matching the input test sample. Our method achieved an average cross-dataset performance of 75.93% Dice index and 67.42% Intersection over Union, surpassing state-of-the-art methods by 3.32% and 3.28%, respectively. Additionally, it achieves real-time execution, processing an average of 29 slices per second due to its significantly reduced number of training parameters—approximately 70% fewer than its closest competitor.