Hierarchical and Robust Pathology Image Reading for High-Throughput Cervical Abnormality Screening

Abstract

Cervical smear screening is an imaging-based cancer detection tool which is of pivotal importance for the early-stage warning. A computer-aided screening system can automatically decide if the images with cervical cells are classified as “abnormal” or “normal”, and then alert pathologists. It can significantly reduce the workload for human experts and is therefore highly-demanded in clinical practice. Most of the screening methods are based on automatic cervical cell detection and classification, but the accuracy is generally limited due to the high variation of cell appearance and lacking of the context information from the surroundings. Here we propose a novel and complete framework for pathology image classification, which can provide a robust screening performance. We commence by implementing the cervical cell detection method to the pathology image from the whole-slide image (WSI) and extract representative patches from the detected “abnormal” cells with corresponding confidence information. The patches are fed into our novel classification model for a more comprehensive analysis, and conduct the classification for the overall target pathology image. It can be demonstrated in experiments that our two-stage method can effectively suppress the errors from cell-level classification, and provide a robust way for pathology image classification.