Maintaining reliable performance in the multi-center deployment of pneumoconiosis diagnosis models is difficult under severe distribution shifts because adaptation must respect the multi-region diagnostic criteria specific to pneumoconiosis. Generic grounded test-time adaptation (TTA) methods are primarily designed for unconstrained classification settings, and their direct application to pneumoconiosis diagnosis can cause semantic drift when there are severe cross-center shifts. In this paper, we propose the ProACT framework, which is a clinically grounded, prototype-guided, anchor-based TTA for stable, multi-center pneumoconiosis diagnosis using fully unlabeled target data. ProACT addresses the challenges of multi-center pneumoconiosis diagnosis using a prototype-guided, anchor-based stabilization mechanism and a sensitivity-aware parameter adaptation strategy. The former uses clinically validated reference radiographs to assess the reliability of the target samples and guide the construction of multiple prototypes, explicitly modeling cross-center intra-class heterogeneity. The latter selectively updates the loss-sensitive components of the feature extractor to preserve the clinically meaningful diagnostic logic. Experiments on a large-scale, real-world multi-center pneumoconiosis benchmark collected from multiple hospitals demonstrate that ProACT consistently outperforms existing deployment-time adaptation methods. ProACT achieves the highest level of agreement with radiologists, and maintains stable performance across different network architectures.
- Article type
- Year
- Co-author
Open Access
Research
Issue
Open Access
Article
Issue
Segmenting greenhouse gases from hyperspectral images can provide detailed information regarding their spatial distribution, which is significant for the monitoring of greenhouse gases. However, accurate segmentation of greenhouse gases is a challenging task due to two main reasons: (1) Diversity: greenhouse gases vary in concentration, size, and texture; (2) Camouflage: the boundaries between greenhouse gases and the surrounding background are blurred. Existing methods primarily focus on designing new modules to address the above challenges, often neglecting the design of the upsampling method within the model, which is crucial for achieving accurate segmentation. In this work, we propose Gas-Aware Upsampling (GasUpper), a novel and efficient upsampling method tailored for greenhouse gas segmentation. Specifically, we first generate a coarse segmentation mask during the upsampling process. Based on the roughly segmented gas and background, we then extract the global features of the gas and combine them with the original features to obtain de-camouflaged feature map that include both the global characteristics of the gas and the local details of the image. This de-camouflaged feature map serves as the foundation for subsequent point sampling. Finally, we utilize the de-camouflaged feature map to generate upsampling coordinate offsets, enabling the model to adaptively adjust the sampling regions based on the content during the sampling process. We conduct comprehensive evaluations by replacing the upsampling method in various segmentation approaches with GasUpper on two hyperspectral datasets. The results indicate that GasUpper consistently and significantly enhances the performance across all segmentation models (0.08%–9.44% Intersection over Union (IoU), 0.47%–6.26% Accuracy), outperforming other upsampling methods.
京公网安备11010802044758号