Annals of Oncology Research and Therapy

REVIEW ARTICLE
Year
: 2021  |  Volume : 1  |  Issue : 1  |  Page : 10--15

A review of automated digital clinical system of breast cancer detection using fine needle aspiration cytology images


Manjula Kalita1, Lipi B Mahanta2, Anup Kumar Das3,  
1 Department of Computer Science, Gauhati University, Guwahati, Assam, India
2 Mathematical and Computational Sciences Division, Institute of Advanced Study in Science and Technology, Guwahati, Assam, India
3 Arya Wellness Centre, Guwahati, Assam, India

Correspondence Address:
Dr. Lipi B Mahanta
Mathematical and Computational Sciences Division, Institute of Advanced Studies in Science and Technology, Vigyan Path, Paschim Boragaon, P.O. Garchuk, Guwahati, Assam
India

Abstract

Screening of microscopic slides is a manual process that involves its subjectivity. A semi-automated computer-based system can contribute to the detection of screening errors by increasing the reliability measure. Traditional machine learning approach or deep learning approach can be used in the semi-automated digital clinical system. The traditional machine learning approach is not very efficient because it involves a lot of heavy mathematics and not able to learn highly complex features. This article presents a systematic summary of the existing solutions of detection of malignancy (breast cancer detection) from fine-needle aspiration cytology images and the segmentation method of nuclei because malignancy can be observed mainly from nuclei feature. It also reports various research issues, challenges and proposes the future research direction. This analysis is helpful for the better use of existing methods and for improving their performance, as well as designing new methods and techniques.



How to cite this article:
Kalita M, Mahanta LB, Das AK. A review of automated digital clinical system of breast cancer detection using fine needle aspiration cytology images.Ann Oncol Res Ther 2021;1:10-15


How to cite this URL:
Kalita M, Mahanta LB, Das AK. A review of automated digital clinical system of breast cancer detection using fine needle aspiration cytology images. Ann Oncol Res Ther [serial online] 2021 [cited 2021 Dec 3 ];1:10-15
Available from: http://www.aort.com/text.asp?2021/1/1/10/322148


Full Text



 Introduction



Breast carcinoma has become the most common malignancy observed in women in developing countries. Breast cancer is curable if detected at an early stage. However, breast cancer is still the leading cause of death due to a lack of adequate facilities for early diagnosis and treatment in low and middle-income countries. Palpable breast lesions can be accurately diagnosed by preoperative tests (such as physical examination, mammography, fine-needle aspiration cytology, and core needle biopsy).[1],[2] The accuracy of diagnosis can be increased by a combination of these preoperative tests. Fine-needle aspiration cytology (FNAC) has become a critical component in the preoperative assessment of breast masses. It has gained popularity due to its fast, easy, and inexpensive approach.[3] The major goal of FNAC is to differentiate benign from malignant lesions.[4],[5] Advancement in artificial intelligence, digital imaging, and computational aid in diagnosis can help to improve the diagnostic accuracy and to reduce the effective workload of a pathologist. In this regard, researchers and practitioners of pathology have been using quantitative analysis for computer-aided diagnosis of pathology samples.[6],[7] The rest of the article is organized as follows. Section 2 presents a brief introduction of breast cancer, discussion on the clinical method (FNAC) is presented in section 3, section 4 presents a model of automated digital clinical system, some of the relevant works are presented in section 5, section 6 presents discussions on related work, section 7 presents Research Issues and Challenges, and section 8 presents the conclusion.

 Breast Cancer



Breast cancer is a disease in which cells in the breast grow out of control and form tumor.[8] However, all the tumors are not cancerous. The tumor can be of two types: Benign or malignant. A benign tumor is a stage when the cells show normal growth, but the production of the cell is higher giving rise to the abnormal lump (compact mass). In the case of a malignant tumor the cell shows abnormal growth, they overgrow in numbers uncontrollably, produce lumps, and result in the cancerous cell. The cells in benign tumor do not spread to other parts of the body from their site of origin, while the cells of a malignant tumor may spread to other parts of the body from their site of origin (metastasize) by the bloodstream or lymphatic system.

Types of breast cancer

Noninvasive breast cancers

Noninvasive breast cancers are cancers that are contained within the milk ducts or lobules in the breast.[9] It is noninvasive because it has not spread into any surrounding breast tissue. Noninvasive cancers are called carcinoma in situ and are sometimes referred to as precancers. Carcinoma in situ is highly treatable because it is a very early cancer, but if it is left untreated or undetected, it can spread into the surrounding breast tissue. The types are:

Ductal carcinoma in situLobular carcinoma in situ.

Invasive breast cancers

The abnormal cancer cells that began forming in the milk ducts or lobules have spread into other parts of the breast tissue.[9] The types are:

Invasive ductal carcinomaInvasive lobular carcinomaPaget's disease of the nippleInflammatory breast cancerPhyllodes tumors of the breastLocally advanced breast cancerMetastatic breast cancer.

 Clinical Method: Fine-Needle Aspiration Cytology



FNAC is a clinical method for the detection of malignancy from breast cytopathology cell samples under a microscope. When a lump or bump is discovered in superficial areas of the body such as the breast and neck, a test, known as FNAC, is recommended to determine whether the lump is cancerous or not. In the FNAC procedure, a small (smaller than those used for blood tests) hypodermic needle is inserted into the lump. The needle is inserted and drawn in and out for about five seconds and then withdrawn. The material is then smeared on a glass slide. This undergoes staining and microscopic examination is done by the pathologist to determine the lump is benign or malignant. [Figure 1] shows some microscopic images (benign and malignant) of pap stained breast's FNAC slide taken by us using Leica ICC50 HD microscope at ×400 resolution using 24 bits color depth and with 5 megapixels camera associated with the microscope.{Figure 1}

Advantages of FNAC test:[3]

Simple and quickMinimally invasive method without unwanted side effectsHighly accurateCost-effective methodPathological assessment of small lesions only.

FNAC is most accurate when experienced cytopathologists are available to test for malignancy and advise on additional aspirations for ancillary tests when needed. Low-grade malignant breast cancers such as mucinous carcinoma may appear clinically and radiologically benign and FNAC plays important role in the correct preoperative diagnosis.[10]

Limitations of fine needle aspiration cytology test

Both FNAC and core needle biopsy (CNB) are equally good in the assessment of breast lesions. However, FNAC is more suitable in developing counties like ours due to better turnaround time and it is cost-effective. CNB can be used as the next step in assessment to minimize the chance of a missed diagnosis of breast cancer. Furthermore, it can be used in cases of low grade or borderline lesions such as atypical ductal hyperplasia, atypical lobular hyperplasia, papillary lesions, tubular carcinoma, and invasive lobular carcinoma, where it is difficult to give a definitive diagnosis on FNAC alone.[11] The main purpose of FNAC or CNB of breast lumps is to confirm cancer preoperatively and to avoid unnecessary surgery in specific benign conditions.

 Automated Digital Clinical System



Digital imaging and computational aid in diagnosis can help to improve the accuracy of the diagnostic procedure and to reduce the effective workload of a pathologist. An automated system can comprise the following steps [Figure 2].{Figure 2}

Preprocessing

The FNAC images acquired using a microscope may be defective and deficient in some respect such as poor contrast and uneven staining, and they need to be improved through the process of image enhancement (part of image preprocessing) which increases the contrast between the foreground (objects of interest) and background.

Segmentation

Image segmentation is the process of partitioning a digital image into regions or categories. The goal of segmentation is to simply change the representation of an image that is more meaningful such that pixels in the same category have similar values; neighboring pixels that are in different categories have dissimilar values. In FNAC images nuclei separation process is very important because the nucleus of the cell is the place where breast cancer malignancy can be observed. Thus, much attention in the construction of the diagnosis system has to be paid to the segmentation stage.

Feature extraction

Feature extraction is a type of dimensionality reduction that efficiently represents the interesting information from the input image as a compact feature vector so that the desired task can be performed by using this reduced representation instead of the complete initial data.

Features that differentiate malignant tumor from benign are:

Large nucleusIrregular size and shape of the nucleusNucleoli are prominent (dark color cell)Tumor malignancy changes concerned with the surface, volume, nucleus/cytoplasm ratio, shape, and density as well as structure and homogeneity.

Classification

Image classification is the process of assigning pixels in the image to categories or classes of interest. Images are classified into benign and malignant class.

Deep learning approach

If we use the deep learning approach it will comprise the feature extraction and classification. Deep neural networks (for example. convolutional neural network [CNN]) can perform feature extraction and classification in one shot. Deep networks can learn highly complex features and as a result, it can improve the diagnosis accuracy.

Machine learning approach

In traditional machine learning approaches, we have to design a feature extraction algorithm that generally involved a lot of heavy mathematics, was not very efficient, and did not perform too well for real-world applications. After doing all of that we would also have to design a classification model to classify the extracted features.

 Related Work



Breast fine needle tumor classification using neural networks

The purpose of this article[12] is to develop an intelligent diagnosis system for breast cancer classification. Artificial neural networks and support vector machines (SVMs) were being developed to classify benign and malignant breast tumor in FNAC. First, the features were extracted from 92 FNAC image. Then these features were presented to several neural network architectures to investigate the most suitable network model for classifying the tumor effectively. Four classification models were used, namely multilayer perceptron (MLP) using back-propagation algorithm, probabilistic neural networks (PNN), learning vector quantization (LVQ), and SVM. The performance of the networks was compared based on resulted error rate, correct rate, sensitivity, and specificity. The method was evaluated using six different datasets including four datasets related to their work and two other benchmark datasets (breast cancer Wisconsin) for comparison. The results showed that the predictive ability of PNNs and SVM is stronger than the others in all evaluated datasets. From their study, they found that more investigations are needed to further improve the clustering algorithms results (FCM, SVM, LVQ, MLP, and PNN) with the selection of different nuclei features and performing hybrid clustering algorithms.

Automated cell nuclei segmentation for breast fine-needle aspiration cytology

In this work,[13] the author presents a fully automated method for microscopic cellular image detection and segmentation. In the first phase, they propose pre-processing techniques due to the low quality of cytological images. This phase includes histogram stretching and contrast-limited adaptive histogram equalization. Next, locations of cell nuclei are detected using circular Hough transform (CHT) and local maximum filtering. Next, for the elimination of false-positive findings that do not correspond to the true nuclei they use Otsu's thresholding and unsupervised fuzzy c-means clustering algorithm. Finally, the locations of the cell nuclei are considered as markers in the watershed transform for the extraction of the nuclei boundaries. From their study, they found that the proposed method can be used as the basis for further processing of cell images, such as the discrimination of normal and abnormal or malignant cells.

Analysis of nuclei textures of fine needle aspirated cytology images for breast cancer diagnosis using complex Daubechies Wavelets

This study[14] states that a Complex Daubechies Wavelet Transform-based decomposition method is applied to breast nuclei in FNAC cytology images and its complex texture features are used with a multivariate classifier. The features extracted through the wavelets are used as input to a k-nn classifier. The results show that the use of the complex wavelet transform gives significantly better results than the ones obtained with real wavelets to classify benign and malignant cases. The author obtained quite satisfactory classification results based on the texture features alone. It is possible that the addition of morphometric features such as nuclear size and shape distributions may improve the classification results still further. It is also possible that more sophisticated classifiers may lead to some improvements.

Segmentation of nuclei in cytological images of breast fine-needle aspiration cytology sample: Case study

The study of different methodologies of cytological image segmentation is proposed here.[15] The study includes the watershed algorithm and active contouring. We can also find here a description of denoising and contrast enhancement techniques; because the raw image taken from a camera mounted on a microscope contains less information and noise. The study covers the different pre-segmentation processes, such as CHT for circle detection and nucleus localization method. Until now many segmentation algorithms were introduced but unfortunately, those cannot be used directly for nuclei segmentation. From the past few years, large efforts are taken to develop a fully automatic segmentation algorithm. Here, a group of modified versions of the cytological image segmentation method adopted for fine needle biopsy images is presented. The consideration of which technique is best for handling a problem of image segmentation highly depends upon the nature of images used for the experiment. In each case, they observed that the problem of connected or overlapping nuclei is still there because it has the same intensity and structure and it is impossible to segment each nucleus.

Morphometric study of nuclei in fine-needle aspiration cytology of breast lesion and its role in the diagnosis of malignancy

The main objectives of this study[16] were (a) To evaluate major axis (MAJX), minor axis (MINX), nuclear area (NA), nuclear perimeter (NP), and nuclear aspect ratio (NAR) using morphometric techniques. (b) To compare these nuclear parameters with their variability in benign and malignant cases and evaluate suitable cut-off values and (c) To study the correlation of these parameters with cytological grades. The authors found that morphometric parameters related to nuclear size and variability evaluated from FNAC material were significantly larger in malignant cases than in benign and they can be gainfully exploited in the diagnosis of breast carcinoma.

High-magnification multi-views based classification of breast fine-needle aspiration cytology cell samples using fusion of decisions from deep convolutional networks

This article[17] presents a deep CNN based classification approach for the diagnosis of the cell samples using their microscopic high-magnification multi-views. An image dataset of 37 (24 benign + 13 malignant) cell sample slides is developed. Manually, multiple regions of interests (ROIs) of size 256 × 256 pixels are selected to represent the cell samples. Overall, 918 ROIs from 175 high magnification views over 37 samples constitute the image dataset. The proposed approach has been tested using the GoogLeNet architecture of CNN. The network was trained using images of 54% cell samples and tested on the rest, achieving 89.7% mean accuracy in 8-fold validation.

Comparative assessment of convolutional neural network architectures for classification of breast fine-needle aspiration cytology images

This article[18] presents a comparison of various deep CNN based fine-tuned transfer learned classification approach for the diagnosis of the FNAC cell samples. The proposed approach has been tested using VGG16, VGG19, ResNet-50, and GoogLeNet-V3 (aka Inception V3) architectures of CNN on an image dataset of 212 images (99 benign and 113 malignant) samples. Later augmented and cleansed to 2120 images (990 benign and 1130 malignant), where the network was trained using images of 80% cell samples and tested on the rest. The authors found that GoogLeNet-V3 is by far the best deep learning method for the classification of FNAC images which gives 96.25% accuracy and it is highly satisfactory.

 Discussions on Related Work



Based on our limited review following observations are made:

The consideration of which technique is best for handling a problem of image segmentation highly depends upon the nature of images used for the experimentAdding more features and adding more sophisticated classifiers may lead to some improvement of existing machine learning approachesThe morphometric parameters related to nuclear size and variability evaluated from FNAC images were significantly larger in malignant cases than in benign and they can be gainfully exploited in the diagnosis of breast carcinomaGoogLeNet-V3 is by far the best deep learning method for FNAC cell image classificationData augmentation by adding more samples and data replication, transfer learning can be used to improve the accuracy of classification.

 Research Issues and Challenges



Image acquisition is a challenging task since benchmark datasets are not easily available. Due to uneven staining of the FNAC slide in the respective laboratory, digitized images may be defective and deficient in contrastDeveloping a single and unified approach for image segmentation that could be used for all sorts of images remains a major challenge. Image segmentation is the key procedure in automating any computer-aided diagnostic system. Accurate segmentation of the region of interest of an image plays a crucial role because it can ultimately determine the success or failure of the computerized carcinoma detection systemMorphological parameters can be used to measure cytological grades of breast carcinoma. Designing an automated system that can detect the cytological grades of breast carcinoma using FNAC images is also a challenging taskThe problem of connected or overlapping nuclei is still there because it has the same intensity and structure and it is impossible to segment each nucleusIn transfer learned classification approach, there is a limitation of the size of datasets. Hence, designing a deep networks model that can work for any size of FNAC dataset is also a challenging taskBreast cancers types are not discussed elaborately in this article because this study is a review on automated systems that can distinguish between benign and malignant tumor, not the types of breast cancer. Designing an automated system that can detect the types of breast cancer is also a challenging task.

 Conclusion



This article presents a comprehensive review of the automated breast cancer detection system from FNAC images. Different authors present different methods for the classification of benign and malignant breast tumor. As malignancy can be detected from the nuclei feature, accurate segmentation of the nuclei from FNAC image is a challenging task and it can help the pathologist to draw some conclusions and to decide the stage of cancer. Therefore, segmentation methods are also discussed here. Finally, we can conclude that deep learning-based methods are more suitable for real-life problem since they can extract more complex features and can give a highly satisfactory result.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.

References

1Ishikawa T, Hamaguchi Y, Tanabe M, Momiyama N, Chishima T, Nakatani Y, et al. False-positive and false-negative cases of fine-needle aspiration cytology for palpable breast lesions. Breast Cancer 2007;14:388-92.
2Bukhari MH, Akhtar ZM. Comparison of accuracy of diagnostic modalities for evaluation of breast cancer with review of literature. Diagn Cytopathol 2009;37:416-24.
3Berner A, Sauer T. Fine-needle aspiration cytology of the breast. Ultrastruct Pathol 2011;35:162-7.
4Mendoza P, Lacambra M, Tan PH, Tse GM. Fine needle aspiration cytology of the breast: The nonmalignant categories. Patholog Res Int 2011;2011:547580.
5Auger M, Hüttner I. Fine-needle aspiration cytology of pleomorphic lobular carcinoma of the breast. Comparison with the classic type. Cancer 1997;81:29-32.
6Demir C, Yener B. Automated Cancer Diagnosis Based on Histopathological İmages: A Systematic Survey. USA, Rensselaer Polytechnic Institute, Department of Computer Science, Technical Report; 2005.
7Irshad H, Veillard A, Roux L, Racoceanu D. Methods for nuclei detection, segmentation, and classification in digital histopathology: A review-current status and future potential. IEEE Rev Biomed Eng 2014;7:97-114.
8Available from: https://www.cdc.gov/cancer/breast/basic_info [Last accessed on 2021 Mar 08].
9Available from: https://www.bcna.org.au. [Last accessed on 2021 Mar 08].
10Adhikari RC, Jha A, Sayami G. Fine needle aspiration cytology findings of mucinous carcinoma of breast: A study of eight cases with histological correlation. J Pathol Nepal 2012;2:285-8.
11Supriya S, Nirmala C, LakshmiS, Raghupathi AR. Fine needle aspiration cytology versus core needle biopsy in breast lessions – A comparative study. Int J Curr Res Rev 2015;7;52-60.
12George YM, Elbagoury BM, Zayed HH, Roushdy MI. Breast fine needle tumor classification using neural networks. Int J Comput Sci Issues 2012;9:247-56.
13George YM, Bagoury BM, Zayed HH, Roushdy MI. Automated cell nuclei segmentation for breast fine needle aspiration cytology. Signal Process 2013;93:2804-16.
14Niwas SI, Palanisamy P, Sujathan K, Bengtsson E. Analysis of nuclei textures of fine needle aspirated cytology images for breast cancer diagnosis using complex daubechies wavelets. Signal Process 2013;93:2828-37.
15Pise AP, Longadge R, Malik LG. Segmentation of nuclei in cytological ımages of breast FNAC sample: Case study. International Journal of Computer Science and Mobile Computing 2014;3:226-32.
16Boruah D, Srinivas V, Belagavi SG. Morphometric study of nuclei in FNAC of breast lesion and its role in diagnosis of malignancy. J Cytol Histol 2014;5:274.
17Garud H, Karri SP, Sheet D, Chatterjee J, Mahadevappa M, Ray AK. High-magnification multi-views based classification of breast fine needle aspiration cytology cell samples using fusion of decisions from deep convolutional networks. IEEE Conference on Computer Vision and Pattern Recognition Workshops 2017;76-81.
18Saikia AR, Bora K, Mahanta LB, Das AK. Comparative assessment of CNN architectures for classification of breast FNAC images. Tissue Cell 2019;57:8-14.