Hybrid AI-assistive diagnostic model permits rapid TBS classification of cervical liquid-based thin-layer cell smears.

Technical advancements significantly improve earlier diagnosis of cervical cancer, but accurate diagnosis is still difficult due to various factors. We develop an artificial intelligence assistive diagnostic solution, AIATBS, to improve cervical liquid-based thin-layer cell smear diagnosis according to clinical TBS criteria. We train AIATBS with >81,000 retrospective samples. It integrates YOLOv3 for target detection, Xception and Patch-based models to boost target classification, and U-net for nucleus segmentation. We integrate XGBoost and a logical decision tree with these models to optimize the parameters given by the learning process, and we develop a complete cervical liquid-based cytology smear TBS diagnostic system which also includes a quality control solution. We validate the optimized system with >34,000 multicenter prospective samples and achieve better sensitivity compared to senior cytologists, yet retain high specificity while achieving a speed of <180s/slide. Our system is adaptive to sample preparation using different standards, staining protocols and scanners.

Breast cancer mitotic cell detection using cascade convolutional neural network with U-Net.

The number of mitotic tumor cells detected in each slide is one of the key indicators of breast cancer prognosis. However, accurate mitotic cell counts are still a difficult problem for pathologists and related experts. Traditional methods use manual design algorithms to extract features of mitotic cells, and most methods rely on sliding windows to achieve pixel-level classification through deep learning. However, the complex background and high resolution of pathological images make the above methods time-consuming and ineffective. In order to solve the above problems, we propose a new cascaded convolutional neural network UBCNN (cascaded CNN based on UNet), which consists of three parts: semantic segmentation and classification to detect mitosis. First, we use an improved UNet ++ segmentation network to locate the candidate set of mitotic targets. Secondly, an adequately labeled cell nucleus data set is sent to an improved two-dimensional VNet network, and the cell nucleus is located by means of semantic segmentation to obtain accurate image blocks of mitotic and non-mitotic cells. Finally, the obtained cell image block is used to train a convolutional neural network to achieve binary classification, and the candidate set area is screened to retain the final result of mitosis cells. This paper verifies the detection effect of the above-mentioned cascade detection algorithm on the ICPR 2012 and 2014 mitosis automatic detection competition data sets. The evaluation indicators include accuracy, recall and F-score. Our cascade detection algorithm based on segmentation and classification reached 0.831 on the ICPR 2012 data set and 0.576 on the ICPR 2014 data set. Compared with other existing algorithms, the detection effect was improved, which was very competitive.