Cytological Images Research Articles

Deep convolutional neural network-based classification of cancer cells on cytological pleural effusion images

Lung cancer is one of the leading causes of cancer-related death worldwide. Cytology plays an important role in the initial evaluation and diagnosis of patients with lung cancer. However, due to the subjectivity of cytopathologists and the region-dependent diagnostic levels, the low consistency of liquid-based cytological diagnosis results in certain proportions of misdiagnoses and missed diagnoses. In this study, we performed a weakly supervised deep learning method for the classification of benign and malignant cells in lung cytological images through a deep convolutional neural network (DCNN). A total of 404 cases of lung cancer cells in effusion cytology specimens from Shanghai Pulmonary Hospital were investigated, in which 266, 78, and 60 cases were used as the training, validation and test sets, respectively. The proposed method was evaluated on 60 whole-slide images (WSIs) of lung cancer pleural effusion specimens. This study showed that the method had an accuracy, sensitivity, and specificity respectively of 91.67%, 87.50% and 94.44% in classifying malignant and benign lesions (or normal). The area under the receiver operating characteristic (ROC) curve (AUC) was 0.9526 (95% confidence interval (CI): 0.9019–9.9909). In contrast, the average accuracies of senior and junior cytopathologists were 98.34% and 83.34%, respectively. The proposed deep learning method will be useful and may assist pathologists with different levels of experience in the diagnosis of cancer cells on cytological pleural effusion images in the future.

Modified U‐Net for cytological medical image segmentation

AbstractDeep learning–based medical image segmentation is henceforth widely established as a powerful segmentation process. This article proposes a new U‐Net architecture based on a convolutional neural network for cytology image segmentation. This structure is more suitable to take into account pixel neighborhood in deconvolution. The goal is to develop an accurate segmentation method for white blood cells segmentation based on cells types features. This new proposed method yields a significant improvement compared to our previous work on the cytological medical dataset. In addition, the performance of the new architecture was also successfully tested on the Digital Retinal Image for Vessel Extraction databases benchmark. The images of this challenge are similar to our cytology image segmentation. Our approach achieved 25% relative improvement of the accuracy compared to the state‐of‐the‐art.

LFANet: Lightweight feature attention network for abnormal cell segmentation in cervical cytology images

With the widely applied computer-aided diagnosis techniques in cervical cancer screening, cell segmentation has become a necessary step to determine the progression of cervical cancer. Traditional manual methods alleviate the dilemma caused by the shortage of medical resources to a certain extent. Unfortunately, with their low segmentation accuracy for abnormal cells, the complex process cannot realize an automatic diagnosis. In addition, various methods on deep learning can automatically extract image features with high accuracy and small error, making artificial intelligence increasingly popular in computer-aided diagnosis. However, they are not suitable for clinical practice because those complicated models would result in more redundant parameters from networks. To address the above problems, a lightweight feature attention network (LFANet), extracting differentially abundant feature information of objects with various resolutions, is proposed in this study. The model can accurately segment both the nucleus and cytoplasm regions in cervical images. Specifically, a lightweight feature extraction module is designed as an encoder to extract abundant features of input images, combining with depth-wise separable convolution, residual connection and attention mechanism. Besides, the feature layer attention module is added to precisely recover pixel location, which employs the global high-level information as a guide for the low-level features, capturing dependencies of channel features. Finally, our LFANet model is evaluated on all four independent datasets. The experimental results demonstrate that compared with other advanced methods, our proposed network achieves state-of-the-art performance with a low computational complexity.

Comparison of Fine-Tuned Deep Convolutional Neural Networks for the Automated Classification of Lung Cancer Cytology Images with Integration of Additional Classifiers.

Objective:It is essential to accurately diagnose and classify histological subtypes into adenocarcinoma (ADC), squamous cell carcinoma (SCC), and small cell lung carcinoma (SCLC) for the appropriate treatment of lung cancer patients. However, improving the accuracy and stability of diagnosis is challenging, especially for non-small cell carcinomas. The purpose of this study was to compare multiple deep convolutional neural network (DCNN) technique with subsequent additional classifiers in terms of accuracy and characteristics in each histology. Methods:Lung cancer cytological images were classified into ADC, SCC, and SCLC with four fine-tuned DCNN models consisting of AlexNet, GoogLeNet (Inception V3), VGG16 and ResNet50 pretrained by natural images in ImageNet database. For more precise classification, the figures of 3 histological probabilities were further applied to subsequent machine learning classifiers using Naïve Bayes (NB), Support vector machine (SVM), Random forest (RF), and Neural network (NN). Results:The classification accuracies of the AlexNet, GoogLeNet, VGG16 and ResNet50 were 74.0%, 66.8%, 76.8% and 74.0%, respectively. Well differentiated typical morphologies were tended to be correctly judged by all four architectures. However, poorly differentiated non-small cell carcinomas lacking typical structures were inclined to be misrecognized in some DCNNs. Regarding the histological types, ADC were best judged by AlexNet and SCC by VGG16. Subsequent machine learning classifiers of NB, SVV, RF, and NN improved overall accuracies to 75.1%, 77.5%, 78.2%, and 78.9%, respectively. Conclusion:Fine-tuning DCNNs in combination with additional classifiers improved classification of cytological diagnosis of lung cancer, although classification bias could be indicated among DCNN architectures.

Classification of Cervical Cytology Overlapping Cell Images with Transfer Learning Architectures

Cervical cell classification is a clinical biomarker in cervical cancer screening at early stages. An accurate and early diagnosis plays a vital role in preventing the cervical cancer. Recently, transfer learning using deep convolutional neural networks; have been deployed in many biomedical applications. The proposed work aims at applying the cutting edge pre-trained networks: AlexNet, ImageNet and Places365, to cervix images to detect the cancer. These pre-trained networks are fine-tuned and retrained for cervical cancer augmented data with benchmark CERVIX93 dataset available publically. The models were evaluated on performance measures viz; accuracy, precision, sensitivity, specificity, F-Score, MCC and kappa score. The results reflect that the AlexNet model is best for cervical cancer prediction with 99.03% accuracy and 0.98 of kappa coefficient showing a perfect agreement. Finally, the significant success rate makes the AlexNet model a useful assistive tool for radiologist and clinicians to detect the cervical cancer from pap-smear cytology images.

Optimization of fuzzy c-means (FCM) clustering in cytology image segmentation using the gray wolf algorithm

BackgroundImage segmentation is considered an important step in image processing. Fuzzy c-means clustering is one of the common methods of image segmentation. However, this method suffers from drawbacks, such as sensitivity to initial values, entrapment in local optima, and the inability to distinguish objects with similar color intensity. This paper proposes the hybrid Fuzzy c-means clustering and Gray wolf optimization for image segmentation to overcome the shortcomings of Fuzzy c-means clustering. The Gray wolf optimization has a high exploration capability in finding the best solution to the problem, which prevents the entrapment of the algorithm in local optima. In this study, breast cytology images were used to validate the methods, and the results of the proposed method were compared to those of c-means clustering.ResultsFCMGWO has performed better than FCM in separating the nucleus from the other dark objects in the cell. The clustering was validated using Vpc, Vpe, Davies-Bouldin, and Calinski Harabasz criteria. The FCM and FCMGWO methods have a significant difference with respect to the Vpc and Vpe indexes. However, there is no significant difference between the performances of the two clustering methods with respect to the Calinski-Harabasz and Davies-Bouldin indices. The results indicate the better efficacy of the proposed method.ConclusionsThe hybrid FCMGWO algorithm distinguishes the cells better in images with less detail than in images with high detail. However, FCM exhibits unacceptable performance in both low- and high-detail images.

STSRNet: Self-Texture Transfer Super-Resolution and Refocusing Network.

Biomedical microscopy images with high-resolution (HR) and axial information can help analysis and diagnosis. However, obtaining such images usually takes more time and economic costs, which makes it impractical in most scenarios. In this paper, we first propose a novel Self-texture Transfer Super-resolution and Refocusing Network (STSRNet) to reconstruct HR multi-focal plane (MFP) images from a single 2D low-resolution (LR) wide field image without relying on scanning or any special devices. The proposed STSRNet consists of three parts: the backbone module for extracting features, the self-texture transfer module for transferring and fusing features, and the flexible reconstruction module for SR and refocusing. Specifically, the self-texture transfer module is designed for images with self-similarity such as cytological images and it searches for similar textures within the image and transfers to help MFP reconstruction. As for reconstruction module, it is composed of multiple pluggable components, each of which is responsible for a specific focal plane, so as to performs SR and refocusing all focal planes at one time to reduce computation. We conduct extensive experiments on cytological images and the experiments show that MFP images reconstructed by STSRNet have richer details in the axial and horizontal directions than input images. At the same time, the reconstructed MFP images also perform better than single 2D wide field images on high-level tasks. The proposed method provides relatively high-quality MFP images when real MFP images cannot be obtained, which greatly expands the application potential of LR wide-field images. To further promote the development of this field, we released our cytology dataset named RSDC for more researchers to use.

A Deep Neural Network for Cervical Cell Detection Based on Cytology Images

A Deep Neural Network for Cervical Cell Detection Based on Cytology Images

DeepCyto: a hybrid framework for cervical cancer classification by using deep feature fusion of cytology images

<abstract> <p>Cervical cancer is the second most commonly seen cancer in women. It affects the cervix portion of the vagina. The most preferred diagnostic test required for screening cervical cancer is the pap smear test. Pap smear is a time-consuming test as it requires detailed analysis by expert cytologists. Cytologists can screen around 100 to 1000 slides depending upon the availability of advanced equipment. Due to this reason Artificial intelligence (AI) based computer-aided diagnosis system for the classification of pap smear images is needed. There are some AI-based solutions proposed in the literature, still an effective and accurate system is under research. In this paper, the deep learning-based hybrid methodology namely DeepCyto is proposed for the classification of pap smear cytology images. The DeepCyto extracts the feature fusion vectors from pre-trained models and passes these to two workflows. Workflow-1 applies principal component analysis and machine learning ensemble to classify the pap smear images. Workflow-2 takes feature fusion vectors as an input and applies an artificial neural network for classification. The experiments are performed on three benchmark datasets namely Herlev, SipakMed, and LBCs. The performance measures of accuracy, precision, recall and F1-score are used to evaluate the effectiveness of the DeepCyto. The experimental results depict that Workflow-2 has given the best performance on all three datasets even with a smaller number of epochs. Also, the performance of the DeepCyto Workflow 2 on multi-cell images of LBCs is better compared to single cell images of other datasets. Thus, DeepCyto is an efficient method for accurate feature extraction as well as pap smear image classification.</p> </abstract>

MULTI-THREAD PARALLELIZING OF CELL CHARACTERISTICS OF BIOMEDICAL IMAGES

An approach to the parallelization of the process of calculating the quantitative characteristics of cell nuclei on biomedical images (cytological, histological, immunohistochemical) is proposed, which will speed up the process of making a diagnosis. The relevance of this task lies in the fact that there are a large number of micro-objects in the image that need to be investigated, and optimization of time and rational distribution of resources will speed up the stage of calculating the area of cell nuclei and their average brightness level. In the future, these data are stored in the database for further use as a dataset for the tasks of classification, clustering, and intellectual analysis. Modern means of data classification and intellectual analysis are used to make a diagnosis. When using convolutional neural networks, the input data to the classifier are images in the format .jpg, .png, .bmp, etc. Alternative algorithms and data processing tools in most cases require quantitative characteristics. In the case of using biomedical images, the quantitative characteristics are the area, perimeter, circumference, length, and major and lateral axes of the cell nucleus. The area and other characteristics of cell nuclei characterize the normal state or the presence of pathologies. Calculating quantitative characteristics on immunohistochemical and cytological images is time-consuming because the number of cell nuclei in one image can be in the range of 10-20 units. To create a dataset of quantitative characteristics of cell nuclei, it is necessary to perform calculations on a large number of images, which in turn requires significant resources, at a particular time. The parallelization of calculating the biomedical image characteristics is implemented on the basis of computer vision algorithms to select the necessary objects and means of software parallelization of tasks at the thread level to speed up the process of calculating the cell nucleus characteristics. It was established that the existing systems of automated microscopy and diagnostic systems based on images do not have the presence of a large number of characteristics of cell nuclei and do not have mechanisms for parallelizing the process of their calculation. The proposed approach makes it possible to speed up the process of calculating the quantitative characteristics of cell nuclei by 25 %. The relevance of the problem of parallelization is due to the need to process a large amount of data for their further reduction and classification. Thread-level parallelization improves image processing speed and does not require specialized hardware.

A Deep Neural Network for Cervical Cell Classification Based on Cytology Images

Cervical cancer is one of the most common cancers among women. Fortunately, cervical cancer is treatable if it is diagnosed timely and administered appropriately. The death rate of cervical cancer has been greatly reduced since Pap smear test was applied. However, Pap smear test is a time-consuming and error-prone process. Moreover, classifying cervical cells into different categories is clinically meaningful but also challenging in the field of cervical cancer detection. To address these concerns, computer-aided diagnosis systems with deep learning need to be designed to automatically analyze cervical cytology images. In this study, we construct a deep convolutional neural network with feature representations learned via multiple kernels with different sizes to automatically classify cervical cytology images, named DeepCELL. Firstly, we design three different basic modules of DeepCELL to capture feature information via multiple kernels with different sizes. Then, we stack several such basic modules to form the cervical cell classification model. Finally, we perform a series of experiments to evaluate the proposed method on two cervical cytology datasets: Herlev and SIPaKMeD. Our method achieves the accuracy of 95.628%, precision of 95.685%, recall of 95.647% and F-score of 95.636% on SIPaKMeD dataset, which are the highest among all competing methods. Similarly, our method also achieves satisfactory result on Herlev dataset. In summary, extensive experimental results demonstrate that our proposed method has a promising performance in cervical cell image classification.

All-in-focus fine needle aspiration biopsy imaging based on Fourier ptychographic microscopy

ContextCytology is the study of whole cells in diagnostic pathology. Unlike standard histologic thinly sliced specimens, cytologic preparations consist of preparations of whole cells where cells commonly cluster and aggregate. As such, cytology preparations are generally much thicker than histologic slides, resulting in large patches of defocus when examined under the microscope. A diagnostic aggregate of cells often cannot be viewed in focus together, requiring pathologists to continually manipulate the focal plane, complicating the task of accurately assessing the entire cellular aggregate and thus in making a diagnosis. Further, it is extremely difficult to acquire useful uniformly in-focus digital images of cytology preparations for applications such as remote diagnostic evaluations and artificial intelligence models. The predominant current method to address this issue is to acquire digital images at multiple focal planes of the entire slide, which demands long scanning time, complex and expensive scanning systems, and huge storage capacity. AimsHere we report a unique imaging method that can acquire cytologic images efficiently and computationally render all-in-focus digital images that are highly compact. Methods and materialThis method applies a metric-based digital refocusing to microscopy data collected with a Fourier ptychographic microscope (FPM). The digitally refocused patches of images are then synthesized into an all-in-focus image. ResultsWe report all-in-focus FPM results of thyroid fine needle aspiration (FNA) cytology samples, demonstrating our method’s ability to overcome the height variance of 30 μm caused by cell aggregation, and rendering images at high resolution (corresponds to a standard microscope with objective NA of 0.75) and that are all-in-focus. ConclusionsThis technology is applicable to standard microscopes, and we believe can have an impact on diagnostic accuracy as well as ease and speed of diagnosing challenging specimens. While we focus on cytology slides here, we anticipate this technology’s advantages will translate well for histology applications. This technique also addresses the issue of remote rapid evaluation of cytology preparations. Finally, we believe that by resolving the focus heterogeneity issues in standard digital images, this technique is a critical advance for applying machine learning to cytology specimens.

A hybrid model for efficient cervical cell classification

Abnormal cell recognition from cervical pap cytology is crucial for early cervical cancer screening and treatment. However, the current manual analysis requires the undivided attention of the pathologists and is prone to diagnostic errors. To tackle this issue, several techniques have been proposed in the last few decades. The success of an automation-based cervical screening and diagnosis system relies on accurate identification of cell types. This paper proposes an efficient automated hybrid framework for enhancing the cell classification accuracy of cervical cytology images. In this work, a Pyramid Scene Parsing Model is used for the segmentation of cell components. The integration of the feature pyramid with local and global context prior makes the model suitable for the segmentation of small cell components. Nuclear as well as cell features of segmented regions are extracted and categorized using an ensemble-based stack classifier. The performance of five different machine learning classifiers is also investigated and it was found that the CNN-based segmentation followed by ensemble-based stack classifier achieved the best performance. The model was evaluated on a publicly available data set and achieved an average accuracy and AUC of 99.7% and 0.996 for 2-class classification and 75.55% and 0.909 for 4-class classification respectively. The experiments demonstrate that the proposed framework achieves promising performance on pap stained cytology images.

Effectiveness of convolutional neural networks in the interpretation of pulmonary cytologic images in endobronchial ultrasound procedures.

BackgroundRapid on‐site cytologic evaluation (ROSE) helps to improve the diagnostic accuracy in endobronchial ultrasound (EBUS) procedures. However, cytologists are seldom available to perform ROSE in many institutions. Recent studies have investigated the application of deep learning in cytologic image analysis. As such, the present study analyzed lung cytologic images obtained by EBUS procedures, and employed deep‐learning methods to distinguish between benign and malignant cells and to semantically segment malignant cells.MethodsNinety‐seven patients who underwent 104 EBUS procedures were enrolled. Four hundred and ninety‐nine lung cytologic images obtained via ROSE, including 425 malignant and 74 benign, and most malignant were lung adenocarcinoma (64.3%). All the images were used to train a residual network model with 101 layers (ResNet101), with suitable hyperparameters selected to classify benign and malignant lung cytologic images. An HRNet model was also employed to mark the area of malignant cells. Automatic patch‐cropping was adopted to facilitate dataset preparation.ResultsMalignant cells were successfully classified by ResNet101 with 98.8% classification accuracy, 98.8% sensitivity, and 98.8% specificity in patch‐based classification; 95.5% classification accuracy in image‐based classification; and 92.9% classification accuracy in patient‐based classification. Malignant cell area was successfully marked by HRNet with a mean intersection over union of 89.2%. The automatic cropping method enabled the system to complete diagnosis within 1 s.ConclusionsThis is the first study to combine lung cytologic image deep‐learning classification with semantic segmentation. The model was optimized for high accuracy and the automatic cropping facilitates the clinical application of our model. The success in both lung cytologic images classification and semantic segmentation on our dataset shows a promising result for clinical application in the future.

The Dynamics of Pathology Dataset Creation Using Urine Cytology as an Example

Introduction: Dataset creation is one of the first tasks required for training AI algorithms but is underestimated in pathology. High-quality data are essential for training algorithms and data should be labelled accurately and include sufficient morphological diversity. The dynamics and challenges of labelling a urine cytology dataset using The Paris System (TPS) criteria are presented. Methods: 2,454 images were labelled by pathologist consensus via video conferencing over a 14-day period. During the labelling sessions, the dynamics of the labelling process were recorded. Quality assurance images were randomly selected from images labelled in previous sessions within this study and randomly distributed throughout new labelling sessions. To assess the effect of time on the labelling process, the labelled set of images was split into 2 groups according to the median relative label time and the time taken to label images and intersession agreement were assessed. Results: Labelling sessions ranged from 24 m 11 s to 41 m 06 s in length, with a median of 33 m 47 s. The majority of the 2,454 images were labelled as benign urothelial cells, with atypical and malignant urothelial cells more sparsely represented. The time taken to label individual images ranged from 1 s to 42 s with a median of 2.9 s. Labelling times differed significantly among categories, with the median label time for the atypical urothelial category being 7.2 s, followed by the malignant urothelial category at 3.8 s and the benign urothelial category at 2.9 s. The overall intersession agreement for quality assurance images was substantial. The level of agreement differed among classes of urothelial cells – benign and malignant urothelial cell classes showed almost perfect agreement and the atypical urothelial cell class showed moderate agreement. Image labelling times seemed to speed up, and there was no evidence of worsening of intersession agreement with session time. Discussion/Conclusion: Important aspects of pathology dataset creation are presented, illustrating the significant resources required for labelling a large dataset. We present evidence that the time taken to categorise urine cytology images varies by diagnosis/class. The known challenges relating to the reproducibility of the AUC (atypical) category in TPS when compared to the NHGUC (benign) or HGUC (malignant) categories is also confirmed.

Comparison of generative adversarial networks architectures forbiomedical images synthesis

The article analyzes and compares the architectures of generativeadversarialnetworks. These networks are based on convolu-tional neural networks that are widely used for classification problems. Convolutional networks require a lot of training data to achieve the desired accuracy. Generativeadversarialnetworks are used for the synthesis of biomedical images in this work. Biomedi-cal images are widely used in medicine, especially in oncology. For diagnosis in oncology biomedical images are divided into three classes: cytological, histological, and immunohistochemical. Initial samples of biomedical images are very small. Getting trainingimages is a challenging and expensive process. A cytological training datasetwas used for the experiments. The article considers the most common architectures of generative adversarialnetworks suchas Deep Convolutional GAN (DCGAN), Wasserstein GAN (WGAN),Wasserstein GAN with gradient penalty (WGAN-GP), Boundary-seeking GAN (BGAN), Boundary equilibrium GAN (BEGAN). A typical GAN network architecture consists of a generator and discriminator. The generator and discriminator are based on the CNN network architecture.The algorithm of deep learning for image synthesis with the help ofgenerativeadversarialnet-worksis analyzed in the work. During the experiments, the following problems were solved. To increase the initial number of train-ingdata to the datasetapplied a set of affine transformations: mapping, paralleltransfer, shift, scaling, etc. Each of the architectures was trainedfor a certain numberof iterations. The selected architectures were compared by the training timeand image quality based on FID(FreshetInception Distance)metric. The experiments were implemented in Python language.Pytorch was used as a machine learning framework. Based on the used softwarea prototype software module for the synthesis of cytological imageswas developed. Synthesis of cytological images was performed on the basis of DCGAN, WGAN, WGAN-GP, BGAN, BEGAN architectures. Goog-le's online environment called Collaboratory was used for the experimentsusing NvidiaTesla K80 graphics processor.

Weakly supervised learning for classification of lung cytological images using attention-based multiple instance learning

In cytological examination, suspicious cells are evaluated regarding malignancy and cancer type. To assist this, we previously proposed an automated method based on supervised learning that classifies cells in lung cytological images as benign or malignant. However, it is often difficult to label all cells. In this study, we developed a weakly supervised method for the classification of benign and malignant lung cells in cytological images using attention-based deep multiple instance learning (AD MIL). Images of lung cytological specimens were divided into small patch images and stored in bags. Each bag was then labeled as benign or malignant, and classification was conducted using AD MIL. The distribution of attention weights was also calculated as a color map to confirm the presence of malignant cells in the image. AD MIL using the AlexNet-like convolutional neural network model showed the best classification performance, with an accuracy of 0.916, which was better than that of supervised learning. In addition, an attention map of the entire image based on the attention weight allowed AD MIL to focus on most malignant cells. Our weakly supervised method automatically classifies cytological images with acceptable accuracy based on supervised learning without complex annotations.

Segmentation of Overlapping Cervical Cells with Mask Region Convolutional Neural Network.

The task of segmenting cytoplasm in cytology images is one of the most challenging tasks in cervix cytological analysis due to the presence of fuzzy and highly overlapping cells. Deep learning-based diagnostic technology has proven to be effective in segmenting complex medical images. We present a two-stage framework based on Mask RCNN to automatically segment overlapping cells. In stage one, candidate cytoplasm bounding boxes are proposed. In stage two, pixel-to-pixel alignment is used to refine the boundary and category classification is also presented. The performance of the proposed method is evaluated on publicly available datasets from ISBI 2014 and 2015. The experimental results demonstrate that our method outperforms other state-of-the-art approaches with DSC 0.92 and FPRp 0.0008 at the DSC threshold of 0.8. Those results indicate that our Mask RCNN-based segmentation method could be effective in cytological analysis.

An ensemble deep learning for automatic prediction of papillary thyroid carcinoma using fine needle aspiration cytology

Accurately cytopathological diagnosis of Papillary Thyroid Carcinoma (PTC) is of importance for reducing costs and increasing efficiency of treatments. In this paper, we pursue that goal by introducing artificial intelligence (AI) for automatic classification of malignant PTC cell clusters from Fine Needle Aspiration Cytology (FNAC) processed by ThinPrep. High-resolution cytological images obtained with a 40 × objective lens digital camera attached to an Olympus microscope were segmented into fragments and then divided into training, validation, and testing subsets. Fragments are non-overlapped patches containing only regions-of-interest that cover informative tissue structures for making proper diagnoses. Deep learning CNN models were pre-trained and fine-tuned on large-scale ImageNet domain before they were re-trained on cytology fragments. Moreover, we proposed a method to compute certainty of the patient-level prediction that undoubtedly provides additional evidence for reliability and confidence of the prediction. Results showed that the best classification performance on digital FNAC images achieved using DenseNet161, obtaining a mean accuracy of 0.9556 (p < 0.01), a mean sensitivity of 0.9734, and a mean specificity of 0.9405 on yet-to-be-seen test-set. Ensemble learning findings suggested combinations of AdaBoost classifier with multiple CNN models boosted predictive performances, up to 0.9971 accuracy. Moreover, stain normalization introduced by Reinhard increased the predictive ability, outperforming histogram specification, and Macenko methods. Presented findings demonstrate deep learning can integrate into computer-aided diagnosis systems to support cytopathologists in accurate diagnosis of PTC.

A novel attention-guided convolutional network for the detection of abnormal cervical cells in cervical cancer screening.

Early detection of abnormal cervical cells in cervical cancer screening increases the chances of timely treatment. But manual detection requires experienced pathologists and is time-consuming and error prone. Previously, some methods have been proposed for automated abnormal cervical cell detection, whose performance yet remained debatable. Here, we develop an attention feature pyramid network (AttFPN) for automatic abnormal cervical cell detection in cervical cytology images to assist pathologists to make a more accurate diagnosis. Our proposed method consists of two main components. First, an attention module mimicking the way pathologists reading a cervical cytology image. It learns what features to emphasize or suppress by refining extracted features effectively. Second, a multi-scale region-based feature fusion network guided by clinical knowledge to fuse the refined features for detecting abnormal cervical cells at different scales. The region proposals in the multi-scale network are designed according to the clinical knowledge about size and shape distribution of real abnormal cervical cells. Our method, trained and validated with 7030 annotated cervical cytology images, performs better than the state of art deep learning-based methods. The overall sensitivity, specificity, accuracy, and AUC of an independent testing dataset with 3970 cervical cytology images is 95.83%, 94.81%, 95.08% and 0.991, respectively, which is comparable to that of an experienced pathologist with 10 years of experience. Besides, we further validated our method on an external dataset with 110 cases and 35,013 images from a different organization, the case-level sensitivity, specificity, accuracy, and AUC is 91.30%, 90.62%, 90.91% and 0.934, respectively. Average diagnostic time of our method is 0.04s per image, which is much quicker than the average time of pathologists (14.83s per image). Thus, our AttFPN is effective and efficient in cervical cancer screening, and improvement of clinical workflows for the benefit of potential patients. Our code is available at https://github.com/cl2227619761/TCT_Detection.