Bandwidth-Efficient Telepathology Viewing Framework via Detector-Guided Hybrid-Resolution Whole-Slide Image Streaming

Whole-slide images (WSIs) are gigapixel, multiresolution pyramids that impose significant bandwidth, latency, and storage demands for interactive remote viewing. It entails a strong fidelity-efficiency trade-off: exhaustive high-resolution tile streaming preserves native details but is Input/Output (I/O) intensive, whereas aggressive compression or view-dependent resampling may improve responsiveness at the cost of visually confounding artifacts. In this work, we present a detectorguided hybrid-resolution viewing framework that transmits only a single low-resolution base level with sparse native ROIs, while reconstructing the complementary non-ROI background to yield a streamable hybrid pyramid. Our base-level tile classifier performs top-down screening to determine multi-level regions of interest (ROIs), whose indices define a determinative cross-resolution coordinate system for accurate alignment and compositing. Non-ROI background is completed by progressive residual-based super-resolution from coarse to fine levels, with hard ROI constraints to preserve diagnostic fidelity. Experimental evaluation on CAMELYON17 dataset with full-resolution WSIs as ground truth, an appropriate base level maintains a tilelevel recall of 0.839 and a node-level recall of 1.000 with a reduction of 99.72% of transmitted data relative to materializing full WSI, and residual background generation reduces nuclei discrepancy by up to 54.7% compared with state-of-the-art methods. These results enable bandwidth-and storage-efficient and diagnostically faithful WSI screening for telepathology.