Contextual analysis and intermediate cell markers enhance high-resolution cell image analysis for automated cervical smear diagnosis

A Review of Computational Methods for Cervical Cells Segmentation and Abnormality Classification

Cervical cancer is the one of the most common cancers in women worldwide, affecting around 570,000 new patients each year. Although there have been great improvements over the years, current screening procedures can still suffer from long and tedious workflows and ambiguities. The increasing interest in the development of computer-aided solutions for cervical cancer screening is to aid with these common practical difficulties, which are especially frequent in the low-income countries where most deaths caused by cervical cancer occur. In this review, an overview of the disease and its current screening procedures is firstly introduced. Furthermore, an in-depth analysis of the most relevant computational methods available on the literature for cervical cells analysis is presented. Particularly, this work focuses on topics related to automated quality assessment, segmentation and classification, including an extensive literature review and respective critical discussion. Since the major goal of this timely review is to support the development of new automated tools that can facilitate cervical screening procedures, this work also provides some considerations regarding the next generation of computer-aided diagnosis systems and future research directions.

Semi-automatic segmentation and classification of Pap smear cells

Cytologic screening has been widely used for detecting the cervical cancers. In this study, a semiautomatic PC-based cellular image analysis system was developed for segmenting nuclear and cytoplasmic contours and for computing morphometric and textual features to train support vector machine (SVM) classifiers to classify four different types of cells and to discriminate dysplastic from normal cells. A software program incorporating function, including image reviewing and standardized denomination of file names, was also designed to facilitate and standardize the workflow of cell analyses. Two experiments were conducted to verify the classification performance. The cross-validation results of the first experiment showed that average accuracies of 97.16% and 98.83%, respectively, for differentiating four different types of cells and in discriminating dysplastic from normal cells have been achieved using salient features (8 for four-cluster and 7 for two-cluster classifiers) selected with SVM recursive feature addition. In the second experiment, 70% (837) of the cell images were used for training and 30% (361) for testing, achieving an accuracy of 96.12% and 98.61% for four-cluster and two-cluster classifiers, respectively. The proposed system provides a feasible and effective tool in evaluating cytologic specimens.

A decision support system to detect morphologic changes of chromatin arrangement in normal-appearing cells

Several studies have described malignancy-associated changes (MACs) of chromatin arrangement in the nuclei of apparently normal cells adjacent to and distant from an invasive cancer area. MAC assessment is a hard task, since it requires a deep knowledge of morphologic features of chromatin arrangement. The aim of this work is to verify the reproducibility of the subjective evaluation of the expert on the basis of a decision support system (DSS) that automatically and objectively reproduces MAC diagnosis. A set of 61 patients with suspected clinical diagnosis for lung cancer has been taken into account. The scientist who first described MAC defined each patient as MAC positive or negative on the basis of the MAC diagnosis performed on all cells of the related cytologic sample. A DSS based on an artificial neural network has been set up to learn the relation between 14 morphometric and texture parameters, computed on each nucleus by image processing techniques, with the MAC diagnosis of the expert on each cell. The results show that an objective automatic assessment on MAC by the DSS can effectively support the MAC diagnosis. The method adopted in this approach may be also appropriate for other problems, where an automatic classification of visually inspected patterns of biological micro- and submicrostructure is needed.

Exploratory analysis of quantitative histopathology of cervical intraepithelial neoplasia: Objectivity, reproducibility, malignancy-associated changes, and human papillomavirus

BACKGROUND

As part of a project to evaluate emerging optical technologies for cervical neoplasia, our group is performing quantitative histopathological analyses of biopsy specimens from 1,190 patients. Objectives in the interim analysis are (a) quantitatively assessing progression of the neoplastic process of cervical intraepithelial neoplasia (CIN)/squamous intraepithelial lesions (SIL), (b) detecting malignancy-associated changes (MACs), and (c) phenotypically measuring human papillomavirus (HPV) detected by DNA testing.

METHODS

The diagnostic region of interest (ROI) from immediately adjacent sections were imaged, and the basal lamina and surface of the superficial layer were delimited. Nonoverlapping quantitatively stained nuclei were selected from 1,190 samples with histopathological characteristics of normal (929), koilocytosis (130), CIN 1 (40), CIN 2 (23), and CIN 3/carcinoma in situ (CIS) (68). A fully automatic procedure located and recorded the center of every nucleus in the region of interest (ROI). We used linear discriminant analysis to assess the changes between normal and CIN 3/CIS.

RESULTS

Scores computed from the cell-by cell features and the clinical grade of CIN/SIL were highly correlated, as were those of the architectural features and the clinical grade of CIN/SIL. We found even higher correlations between a combination of cell-by-cell and architectural scores, and clinical grade. Using these scores, we found MACs in the normal biopsy specimens from patients with high-grade CIN/SIL. Furthermore, the same scores correlated with the molecular detection of HPV.

CONCLUSIONS

Quantitative histopathology can be used in large clinical trials as an objective and reproducible measure of CIN/SIL. Detectable phenotypic changes correlate well with CIN/SIL neoplastic progression. It can also be used to infer the presence of CIN/SIL (MACs) and molecular changes associated with increased risk of cancer development (high-risk HPV).

Optically computing the hit-miss transform for an automated cervical smear screening system

The hit-miss transform serves as a region-of-interest locator for cells from cervical smear images that show abnormal changes, which are indicative of malignancy, in their nuclei. An optical implementation of the hit-miss transform algorithm uses an analog spatial light modulator for gray-scale modulation at the filter plane of a 4f optical correlator. Gray-scale modulation at the filter plane improves correlator performance in comparison with a binary phase-only filter (BPOF) by reduction of the edge enhancement of kernels used in morphological detection of cancerous cervical cells. The hit-miss transform with a gray-scale amplitude and binary phase optical filter (GABPOF) for the hit filter and a BPOF for the miss filter shows a 47% reduction in total error versus the use of only BPOF filters to locate abnormal cells.

Automated screening of cervical cytology specimens

After more than four decades of research into automation of the process of screening Papanicolaou (Pap) smears, attempts to develop commercially viable automated screening machines have increased in recent years. These developments have been made possible in part because of the improving price-to-performance ratios in computers and other electronics. Although the Pap smear has been responsible for a very significant decrease in mortality of cervical cancer over the past 40 years, concern has arisen over false-negative cases, with their effects on patients, and the associated legal liability, particularly in the United States. In addition, shortages of cytotechnologists, which have been exacerbated by new regulations limiting the number of slides that may be examined per day, have caused concern about handling the workload, which will probably increase as more individuals gain access to preventive health care. Automated screening machines can potentially allow detection of abnormal cases that are missed with conventional screening, although they may substantially increase the cost of Pap smears. The use of automated screening machines represents a change in the way cervical cytology specimens are processed, and with some machines, a significant change in the operation of the cytology laboratory. Current methods for processing and evaluating Pap smears have not changed significantly for the past four decades. This review discusses some of the principles of operation and practical aspects of automated screening machines.

Quantitative evaluation of malignant potential of early breast cancer using high resolution image cytometry

We have been testing two independent hypotheses which postulate that the malignant potential of early breast cancer can be objectively assessed by measuring nuclear features of the diagnostic-malignant cells (hypothesis I) and/or normal-appearing epithelial cells found in the vicinity of the carcinoma (hypothesis II). In preliminary experiments, we tested some of these hypotheses using historical samples and a high resolution image cytometry apparatus. Tissue sections were stained with our stoichiometric stain and over 60 nuclear features, primarily texture features describing the DNA distribution in the nuclei, were employed in the multivariate analyses. Data derived from measurements of ductal carcinoma in situ (DCIS) with and without the invasive component indicated that the malignant potential of these lesions can be estimated with a sensitivity and specificity of at least 80%. The analysis of the tissue surrounding an invasive breast carcinoma showed that the existing malignancy can be predicted solely from the measurements of normal nuclei (normal-appearing breast lobules) in more than 85% of patients. This result indicates that the analysis of benign tissues also could give prognostically valid information. These results can be greatly improved using larger sample sizes and other improvements, including technical improvements of the cytometry device. We believe that this approach can be developed into a practical diagnostic and prognostic tool for better management of early breast cancer.