CC BY
1*
ALT'23 The 30th International Conference on Advanced Laser Technologies
B-I-28
A digital pathology technique based on Mueller matrix microscopy
Jiachen Wan, Haojie Pei, Yue Yao and Hui Ma
Shenzhen International Graduate School, Tsinghua University, 518055, China Email: mahui@tsinghua.edu.cn
Digital pathology aims to assist pathologists during diagnostic process through the digitization of pathological slides followed by extraction of the diagnostic information. Mueller matrix microscopy is an emerging optical imaging method that is label-free and non-invasive, capable of revealing microstructural details at subcellular scale. It provides invaluable diagnostic information to pathologists that is otherwise unavailable through other non-polarization optical microscopy techniques.
Here we report a label-spreading method based on evaluating super-pixels from Mueller matrix microscopic images to represent their polarization features and propagating the pathologist's initial manual label of cancerous region to the entire field of view in finer detail, highlighting regions that share the same microstructural characteristic with pathologist's labeled region. A human-in-the-loop design is adopted which allows the pathologists to play a crucial role in supervising the label-spreading process by controlling essential model parameters and providing feedback on the label-spreading quality. After sufficient iterations, the label-spreading technique predicts all the potential candidates of cancerous regions and leads to substantial reduction in the diagnostic workload of doctors. In the meantime, the label-spreading process generates a vast amount of high-quality labeled image patches that will serve as invaluable data for other downstream tasks, particularly deep learning feature extractions of high resolution images in whole slide imaging (WSI). This technique is a key step towards realizing the assisted diagnostic application based on Mueller matrix microscopy, expanding the possibilities for other prospective applications in the future.
