A novel automation-assisted cervical cancer reading method based on convolutional neural network

Abstract

Cervical cytology screening using Pap smear or liquid-based cytology is one of the most widely followed and accepted method. Automation-assisted screening based on cervical cytology has become a necessity due to the manual screening method operated by a visual analysis for cervical cell specimen under the microscope of the glass slide is usually labor-intensive and time-consuming. 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 1,014 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.

Using Dynamic Features for Automatic Cervical Precancer Detection

“…There are several techniques for computer-aided diagnosis of cervical cancer [ 7 , 8 , 9 , 10 ]. Most of them use cytological images [ 11 , 12 , 13 , 14 , 15 , 16 , 17 ]. However, an increasing number of studies are developing methods for automatic classification of images captured during VIA, and often, adding images taken during the visual inspection with Lugol’s iodine (VILI) or with the green lens.…”

Mutation based Atom Search Optimization Algorithm for Hyperparameter Optimization of the DenseNet121 Architecture for Cervical Cancer Classification

“…The benchmark Herlev datasets along with liquid-based cytology and conventional Pap smear methods are used in the experimental evaluation. Xiang et al [12] suggested a totally segmentation-free method and convolutional neural network (CNN) without specific hand-crafted feature designs. Four highly similar categories are introduced to enhance the classification performances.…”

“…Confusion matrix received from MASO optimized DenseNet 121 modelThe state-of-art comparison results based on cervical cancer detection are depicted in Table7. We have selected five state-of-art methods such as the proposed MASO optimized DenseNet 121method with support vector machine[7], Enhanced Fuzzy C-Means (EFC-means) algorithm[9], Fully Convolutional Neural Network (FCN)[11], and CNN[12] methods. The proposed method achieved 98.38% accuracy,98.5%specificity,98.83% sensitivity,98.58% precision,99.3%recall and 98.25% F-score.…”

Segmentação de Núcleos Celulares Baseada em Agrupamento e Características de Forma

“…O trabalho de [Xiang et al 2020] propõe um método diferente com o uso de redes neurais convolucionais que não conta com a etapa de segmentac ¸ão, como é mais comum em trabalhos nesse segmento. O método conta com a rede YOLOv3 [Redmon and Farhadi 2018] como base para a detecc ¸ão de células e aglomerados como objetos alvo e sem focar na extrac ¸ão manual de características, atingindo resultados de 0.975 e 0.678 de sensitividade e especificidade, respectivamente na classificac ¸ão em 10 tipos diferentes de células.…”