Introduction: Pneumonia is one of the deadliest lower respiratory tract infections worldwide, and rapidly distinguishing bacterial pneumonia, viral pneumonia, and normal chest radiographs is crucial for guiding precise medication and avoiding misdiagnosis. However, the imaging manifestations of bacterial and viral pneumonia on chest X-rays often overlap, and some normal chest X-rays may also show atypical changes. Clinical differentiation relies heavily on the expertise of senior physicians. In primary care or emergency settings where experienced radiologists are scarce, this can readily lead to diagnostic delays and antibiotic misuse, further exacerbating the risk of antimicrobial resistance and severe outcomes. Methods: To address this challenge, this study proposes a lightweight network named FLIFuse. By adopting a parallel multi-branch architecture, the model thoroughly mines discriminative imaging features, substantially reducing computational overhead while maintaining high classification accuracy. This provides a technical solution that balances accuracy and efficiency for medical image analysis in resource-constrained environments. FLIFuse consists of three heterogeneous branches: an ExtraDW module that enhances local texture details, a Conv2Next module that captures global semantic structures, and an FLI Block that fuses multi-granularity information across different receptive fields to strengthen the response in critical regions. Results: The model occupies only 33.43 M parameters and incurs a computational cost of 6.61 GFLOPs. On the public ChestXRay2017 dataset, FLIFuse achieves a three-class classification accuracy of 91.4% for normal, bacterial pneumonia, and viral pneumonia, outperforming several current state-of-the-art models and attaining extremely low resource consumption while ensuring diagnostic accuracy. In cross-domain tests on an external COVID-19 chest X-ray dataset and the Nerthus endoscopic image dataset, the model also exhibits stable transferability, demonstrating its potential for cross-domain adaptation across different medical imaging tasks. Discussion: These results demonstrate that FLIFuse strikes a favorable balance between diagnostic accuracy and computational efficiency, providing an efficient and lightweight model option for computer-aided interpretation of pneumonia imaging, and showing potential for translation into clinical auxiliary diagnosis.

Pulmonary lesion segmentation, a key component of computer-aided diagnosis, has significant clinical value in accurately localizing infected regions, quantifying disease progression, and guiding individualized treatment plans. To address the limitations of traditional methods—reliance on handcrafted features, vulnerability to medical image noise, and susceptibility to physician subjectivity—this study proposes HFSUNet, an improved UNet-based segmentation model. The model introduces the multi-scale attention module (MSAM) module to capture cross-scale salient lesion features (alleviating boundary blurring and detail loss), Haar wavelet downsampling (HWD) to reduce downsampling information loss, and marginal weight loss (MWL) to optimize edge segmentation affected by blurred lesion boundaries. Experimental results show HFSUNet achieves mIoU scores of 99.75% (COVID-19), 86.91% (MosMedData), and 92.77%, 92.28%, and 94.16% on DDTI, BUSI, and TN3K, outperforming UNet-based baselines. Notably, while maintaining high segmentation accuracy, it only has 5.638 million parameters and 11.582 GFLOPs, significantly reducing complexity and computational cost. This advantage enables clinical application in resource-limited settings, demonstrating great potential in lesion identification, diagnostic assistance, and prognosis support.