基于不确定性机制的病理图像病变区域的语义分割方法

Efficient Semantic Segmentation Method of Lesion Area in Pathology Images Based on Uncertainty Mechanism

随着病理图像的数字化和计算机深度学习技术的发展,使用深度学习技术处理分析数字化病理图像对缩短诊断流程和提高临床效率有重大的实际意义。然而,数字化病理图像的分辨率极高,传统的深度学习网络和图像处理技术往往需要付出很高的硬件和时间成本。文章引入滑窗机制和多尺度信息处理高分辨率的数字化病理图像,同时将多尺度信息与不确定性机制相结合,极大地提高了处理效率,将平均处理时间降到1min以内。

With the digitization of pathological images and advancements in deep learning technology,using deep learning to process high-resolution digital pathology images is crucial for expediting diagnosis and enhancing clinical efficiency.Traditional deep learning networks and image processing methods are often resource-intensive,given the high resolution of these images.This paper introduces a novel approach,combining a sliding window mechanism with multi-scale information and an uncertainty mechanism,significantly improving efficiency.This innovation reduces average processing time to within 1 min.