Pixel-level annotation of lung cavities (LCs) in computed tomography (CT) images is challenging due to their morphological diversity and complexity. Weakly supervised semantic segmentation (WSSS) methods, which utilize sparse annotations (e.g., image-level labels), offer a promising solution. However, existing WSSS approaches often generate coarse pseudo-labels and lack sufficient spatial supervision, resulting in under- or over-segmentation of irregular lesions. To address these limitations, we introduce several key innovations. First, we propose a novel Graph-based Affinity Network (GA-Net) that, unlike conventional methods relying on low-level pixel features, models long-range contextual relationships and structural dependencies using a superpixel graph and learned edge inference kernel, enabling structure-aware pseudo-label refinement for complex lesion morphology. Second, we introduce region-wise affinity propagation, which refines segmentation by propagating activations within semantically coherent 3D regions, offering more precise control over under-/over-segmentation compared to global affinity methods. Additionally, we incorporate Exponential Moving Average (EMA) ensembling for training stability and a scribble-based segmentation module that utilizes pseudo-label contours to provide direct boundary supervision. Extensive experiments on three benchmark datasets demonstrate that our method outperforms existing state-of-the-art medical WSSS techniques, achieving precise and reliable segmentation of complex LCs in CT scans.
Tuberculosis (TB) remains a significant global health challenge, characterized by high incidence and mortality rates on a global scale. With the rapid advancement of computer-aided diagnosis (CAD) tools in recent years, CAD has assumed an increasingly crucial role in supporting TB diagnosis. Nonetheless, the development of CAD for TB diagnosis heavily relies on well-annotated computerized tomography (CT) datasets. Currently, the available annotations in TB CT datasets are still limited, which in turn restricts the development of CAD tools for TB diagnosis to some extent. To address this limitation, we introduce DeepPulmoTB, a CT multi-task learning dataset explicitly designed for TB diagnosis. To demonstrate the advantages of DeepPulmoTB, we propose a novel multi-task learning model, DeepPulmoTBNet (DPTBNet), for the joint segmentation and classification of lesion tissues in CT images. The architecture of DPTBNet comprises two subnets: SwinUnetR for the segmentation task, and a lightweight multi-scale network for the classification task. Furthermore, to enhance the model's capacity to capture TB lesion features, we introduce an improved iterative optimization algorithm that refines feature maps by integrating probability maps obtained in previous iterations. Extensive experiments validate the effectiveness of DPTBNet and the practicality of the DeepPulmoTB dataset.