Search

Background: Although many cervical cytology diagnostic support systems have been developed, it is challenging to classify overlapping cell clusters with a variety of patterns in the same way that humans do. In this study, we developed a fast and accurate system for the detection and classification of atypical cell clusters by using a two‐step algorithm based on two different deep learning algorithms. Methods: We created 919 cell images from liquid‐based cervical cytological samples collected at Sapporo Medical University and annotated them based on the Bethesda system as a dataset for machine learning. Most of the images captured overlapping and crowded cells, and images were oversampled by digital processing. The detection system consists of two steps: (1) detection of atypical cells using You Only Look Once v4 (YOLOv4) and (2) classification of the detected cells using ResNeSt. A label smoothing algorithm was used for the dataset in the second classification step. This method annotates multiple correct classes from a single cell image with a smooth probability distribution. Results: The first step, cell detection by YOLOv4, was able to detect all atypical cells above ASC‐US without any observed false negatives. The detected cell images were then analyzed in the second step, cell classification by the ResNeSt algorithm, which exhibited average accuracy and F‐measure values of 90.5% and 70.5%, respectively. The oversampling of the training image and label smoothing algorithm contributed to the improvement of the system's accuracy. Conclusion: This system combines two deep learning algorithms to enable accurate detection and classification of cell clusters based on the Bethesda system, which has been difficult to achieve in the past. We will conduct further research and development of this system as a platform for augmented reality microscopes for cytological diagnosis. [ABSTRACT FROM AUTHOR]

While automation-assisted reading system can improve efficiency, their performance often relies on the success of accurate cell segmentation and hand-craft feature extraction. This paper presents an efficient and totally segmentation-free method for automated cervical cell screening that utilizes modern object detector to directly detect cervical cells or clumps, without the design of specific hand-crafted feature. Specifically, we use the state-of-the-art CNN-based object detection methods, YOLOv3, as our baseline model. In order to improve the classification performance of hard examples which are four highly similar categories, we cascade an additional task-specific classifier. We also investigate the presence of unreliable annotations and coped with them by smoothing the distribution of noisy labels. We comprehensively evaluate our methods on our test set which is consisted of 1014 annotated cervical cell images with size of 4000 × 3000 and complex cellular situation corresponding to 10 categories. Our model achieves 97.5% sensitivity (Sens) and 67.8% specificity (Spec) on cervical cell image-level screening. Moreover, we obtain a best mean average precision (mAP) of 63.4% on cervical cell-level diagnosis, and improve the average precision (AP) of hard examples which are the most valuable but most difficult to distinguish. Our automation-assisted cervical cell reading system not only achieves cervical cell image-level classification but also provides more detailed location and category reference information of abnormal cells. The results indicate feasible performance of our method, together with the efficiency and robustness, providing a new idea for future development of computer-assisted reading systems in clinical cervical screening. [ABSTRACT FROM AUTHOR]