Abstract:The diagnosis of squamous cell carcinoma requires the accurate classification of cervical squamous lesions in the ThinPrep cytologic test (TCT). It primarily relies on a pathologist's interpretation under a microscope. Deep convolutional neural networks (DCNN) have played an increasingly important role in digital pathology. However, they have not been applied to diverse datasets and externally validated. In the present study, a DCNN model based on VGG16 and an ensemble training strategy (ETS) based on 5-fold cross-validation was employed to automatically classify normal and abnormal cervical squamous cells from a multi-center dataset. First, we collected a dataset comprising 82 TCT samples from four hospitals and fine-tuned our model twice on the dataset with and without the ETS. Then, we compared the classifications obtained from the models with those provided by two skilled pathologists to discriminate the performance of the models in terms of classification accuracy and efficiency. Finally, paired sample t-tests were used to validate the consistency between the classification provided by the proposed methods and that of the pathologists. The results showed that ETS slightly, though not significantly, improved the classification accuracy compared with that of the pathologists: P0=0.387>0.05 (DCNN without ETS vs. DCNN with ETS), P1=0.771>0.05 (DCNN with ETS vs. pathologist 1), P2=0.489>0.05 (DCNN with ETS vs. pathologist 2). The DCNN model was almost 6-fold faster than that of the pathologists. The accuracy of our automated scheme was similar to that of the pathologists, but a higher efficiency in the accurate identification of cervical squamous lesions was provided by the scheme. This result allows for wider and more efficient screening and may provide a replacement for pathologists in the future. Future research should address the viability of the practical implementation of such DCNN models in the laboratory setting.

DSCC: a Data Set of Cervical Cell Images for Cervical Cytology Screening

Artificial Classification of Cervical Squamous Lesions in ThinPrep Cytologic Tests Using a Deep Convolutional Neural Network.

TSO-DETR: A Network for Small Object Detection of Cervical Cells in TCT Smear

DCCL: A Benchmark for Cervical Cytology Analysis

Cx22: A new publicly available dataset for deep learning-based segmentation of cervical cytology images

A Cervical Histopathology Dataset for Computer Aided Diagnosis of Precancerous Lesions

BCData: A Large-Scale Dataset and Benchmark for Cell Detection and Counting

A New Cervical Cytology Dataset for Nucleus Detection and Image Classification (Cervix93) and Methods for Cervical Nucleus Detection

MSCCNet: Multi-Scale Convolution-Capsule Network for Cervical Cell Classification

CAISHI: A benchmark histopathological H&E image dataset for cervical adenocarcinoma in situ identification, retrieval and few-shot learning evaluation

Annotated Pap cell images and smear slices for cell classification

AI-Driven Cervical Cancer Cytological Diagnosis Solution based on Large Scale Data Collections and Annotations: A Multi-centre Clinical Validation

SCAC: A Semi-Supervised Learning Approach for Cervical Abnormal Cell Detection.

A Survey on Deep Learning-based Cervical Cytology Screening: from Cell Identification to Whole Slide Image Analysis

Selective Detection and Segmentation of Cervical Cells

CytoBrain: Cervical Cancer Screening System Based on Deep Learning Technology

Discrimination of Cervical Cancer Cells Via Cognition-Based Features

An automatic segmentation method for multispectral microscopic cervical cell images

Hierarchical pathology screening for cervical abnormality

Channel Spatial Collaborative Attention Network for Fine-Grained Classification of Cervical Cells.

Quantitative analysis of abnormalities in gynecologic cytopathology with deep learning