Pre-cancer risk assessment in habitual smokers from DIC images of oral exfoliative cells using active contour and SVM analysis
Introduction
Oral cancer is one of the most commonly occurring form of cancer disease in the Indian sub-continent and it ranks amongst the top three cancers in the world (Globocan 2012, 2014). Among several factors, smoking is known to contribute significantly to the development of oral cancer (Rodgmanand and Perfetti, 2013). Advances in clinical research has led to development of early detection processes which facilitate wide treatment options and enhance the survival rate (Ford and Farah, 2013). However, traditional non-invasive detection methods generally suffer from low specificity and often produces false positive results (Messadi, 2013, Nair et al., 2012). Moreover, manual evaluation of microscopic images is subject to variation in perceptions and level of expertise of the pathologists (Mehrotra et al., 2011). To overcome the subjectivity concerns, computer assisted morphometric analysis has been adopted to measure the cytological alterations as indication of cellular status and aid in the process of early cancer diagnosis (Shin et al., 2010). For non-invasive assessment, phase contrast (unstained) images have been previously used for cytological assessment (Naadaf et al., 2014). However, the phase contrast images provide only two dimensional information and such images often suffer from ‘halo’ effect, which is nothing but a false bright areas or reverse contrast around phase objects (Davidson et al., 2015). Automated processing of these images becomes challenging due to the low contrast between the foreground objects (cells) and the background (Jaccard et al., 2014).
Differential interference contrast (DIC) microscopy images, on the other hand, permit effective optical sectioning with sharp boundary features with detailed resolution of cell morphology (Davidson et al., 2015, Chena et al., 2013). However, to the best of our knowledge, processing of DIC images has still not extensively developed or reported techniques are yet not optimized as processed images often exhibit over-segmentation of image objects (Kuijper and Heise, 2008, Obara et al., 2013). Accuracy in segmentation is critical to improvement in image analysis performance captured from diverse modalities. Active contour or snake model is widely applied in segmentation of optical coherence tomography (OCT) or images obtained through magnetic resonance imaging (MRI) with satisfactory results (Boonthong et al., 2014, Artan et al., 2011). However, not much reports are available on application of active contour method for segmentation of optical microscopic images (Hamarneh et al., 2000). Present study explores the possibility of segmentation of DIC images of oral epithelial cells through active contour snake model to distinguish cancer cells from normal.
On the other hand, different analytical methods are used for prediction of pre-malignant trend, based on sequential changes in cellular and nuclear levels among the habitual smokers (Noroozi and Zakerolhosseini, 2016). However, according to the literature very few automated systems are available for oral cancer detection based on feature extraction (Krishna et al., 2012, Krishna et al., 2010). To obtain the prediction accuracy, various morphological features are analyzed for evaluating cellular abnormalities (Domcke et al., 2013). Beside the general morphological features, namely cellular area (CA), nuclear area (NA), nucleus–cytoplasmic ratio (N:C), etc., present study incorporates fifteen additional features, from both cell and nucleus of oral epithelia for improving the accuracy of pre-cancer trend analysis. Since the DIC images provide textural information which are important for grading different phases of pathological condition, the texture features of DIC images using a second order texture calculation method, gray level co-occurrence matrix (GLCM) has been examined for better prediction control (Anuradha and Sankaranarayanan, 2013). GLCM is the description of relationship of two neighbouring pixels (Mohanaiah et al., 2013).
For class separation among the different groups, under study investigation with the help of these above mentioned extracted features, SVM based learning method has been adopted for classification (Rathore et al., 2015). This supervised learning method has been reported previously to address many cancer classification problems for its capability to handle n-dimensional feature space and less execution time (Ang et al., 2015, Chinnu, 2015, Amlica et al., 2015). Recent studies on classification of oral cancer using SVM based on textural features has provided 93% accuracy (Anuradha and Sankaranarayanan, 2013). In 2015, Sharma et al. Sharma and Om (2015) had used SVM based classification method for oral cancer pattern extraction from histopathological images.
Present study thus proposes a novel screening method for pre-cancer risk assessment, from DIC images of epithelial cells for habitual smokers. Active contour snake model is used for segmentation of DIC images. Cellular as well as nuclear features are then extracted for risk prediction/early cancer trend assessment using SVM method.
The remainder of the paper is organized as follows. Scope and contribution of the present work is presented in Section 2. Section 3 describes the mathematical preliminaries, tools, techniques and measures used for this study. Proposed method of data acquisition is described in Section 4. Experimental results and performance evaluation of the proposed method are described in Section 6. Finally, Section 7 concludes the paper along with the scope of the future works.
Section snippets
Scope and contributions of the present work
An automatic oral pre-cancer screening methodology as an aid in early diagnosis followed by improvement in survival rate is of utmost importance. To this aim, a method is suggested with the following scopes and contributions:
- •
The study is based on a non-invasive assessment and direct analysis of clinical samples. Some literatures have suggested the use of phase contrast images for early detection of oral cancer. But due to the presence of ‘halo’ effect, it is difficult to identify the clear
Mathematical preliminaries: tools, techniques and measures
This section describes briefly various mathematical tools and techniques used in this work with an objective to make the article self-contained.
Acquisition development
This section describes the procedure adopted for patients selection, sample collection, preparation of microscopic samples and its acquisition for this study.
Proposed method
Different steps for the proposed method, in an integrated form, are represented by schematic diagram in Fig. 1 that include tools, techniques used, image acquisition, processing with feature extraction and classification. Their operations are described as follows:
Step 1: Image Acquisition: Ethanol fixed slides of oral epithelial samples of five study groups were assessed through DIC mode of Nikon inverted microscope (Nikon eclipse T, Japan) at 20× optical magnification. For PAP stained slides,
Experimental results and discussion
This section describes the experimental outcome, contributions of different techniques and performance metrics considered for the proposed system.
Conclusion and scope of future works
Present study demonstrates that the risk assessment through SVM classifier based on cyto-morphological features have potential to predict risk amongst habitual smokers for developing oral cancer. The study concludes that label free monitoring and non-invasive diagnosis system is effective in predicting cancer risk development. Segmentation of DIC images of oral exfoliative cells for cancer detection followed by active contour method was found to improve results. Although present study reported
Conflict of interest
None declared.
Acknowledgements
The authors thank Dr. M. Muthu Rama Krishnan, TEQIP-II and DST Fast Track for their contribution in computation aspect of the study.
References (52)
- et al.
Early detection and diagnosis of oral cancer: strategies for improvement
J. Cancer Policy
(2013) - et al.
Differential diagnosis of squamous cell carcinoma in situ using skin histopathological images
Comput. Biol. Med.
(2016) - et al.
Automated colon cancer detection using hybrid of novel geometric features and some traditional features
Comput. Biol. Med.
(2015) - et al.
A novel generalized gradient vector flow snake model using minimal surface and component-normalized method for medical image segmentation
Biomed. Signal Process. Control
(2016) Estimated Cancer Incidence, Mortality and Prevalence Worldwide in 2012
(2014)- et al.
The Chemical Components of Tobacco and Tobacco Smoke
(2013) Diagnostic aids for detection of oral precancerous conditions
Int. J. Oral Sci.
(2013)- et al.
Oral cancer: premalignant conditions and screening – an update
J. Cancer Res. Therap.
(2012) - et al.
The efficacy of oral brush biopsy with computer-assisted analysis in identifying precancerous and cancerous lesions
Head Neck Oncol.
(2011) - et al.
Advances in fluorescence imaging techniques to detect oral cancer and its precursors
Future Oncol.
(2010)
A phase contrast cytomorphometric study of squames of normal oral mucosa and oral leukoplakia: original study
J. Oral Maxillofac. Pathol.
Comparison of Phase Contrast and DIC Microscopy. Tech. Rep.
Trainable Segmentation of Phase Contrast Microscopy Images Based on Local Basic Image Features Histograms (Document)
Complex local phase based subjective surfaces (CLAPSS) and its application to DIC red blood cell image segmentation
Neurocomputing
An automatic cell segmentation method for differential interference contrast microscopy
Bacterial cell identification in differential interference contrast microscopy images
BMC Bioinformatics
Semi-automated detection of breast mass spiculation using active contour
Signal and Information Processing Association Annual Summit and Conference (APSIPA)
Graph-based active contours using shape priors for prostate segmentation with MRI
Active contour models: application to oral lesion detection in color images
IEEE Conf. Syst. Man Cybern.
Automated oral cancer identification using histopathological images: a hybrid feature extraction paradigm
Micron
Automated characterisation of sub-epithelial connective tissue cells of normal oral mucosa: Bayesian approach
Evaluating cell lines as tumour models by comparison of genomic profiles
Nat. Commun.
Statistical feature extraction to classify oral cancers
J. Glob. Res. Comput. Sci.
Image texture feature extraction using GLCM approach
Int. J. Sci. Res. Publ.
MRI brain tumor classification using SVM and histogram based image segmentation
Int. J. Comput. Sci. Inf. Technol.
Cited by (20)
Computational analysis of histological images from hematoxylin and eosin-stained oral epithelial dysplasia tissue sections
2022, Expert Systems with ApplicationsFeature assisted cervical cancer screening through DIC cell images
2021, Biocybernetics and Biomedical EngineeringCitation Excerpt :In digital pathology, quantitative microscopy with a computer-aided diagnosis (CAD) system has been used to identify and analyze abnormalities in an early stage of cancer. Recent studies demonstrate application of label-free, quantitative methods like phase contrast (PhC) and differential interference contrast (DIC) microscopy for lungs [3], cervical [4], oral cancer detection [5]. In comparison to the PhC approach, the DIC method provides more detailed information and a pseudo-three-dimensional representation of cells, making it easier to identify cellular morphological changes.
What is the state of the art of computer vision-assisted cytology? A Systematic Literature Review
2021, Computerized Medical Imaging and GraphicsInferential Survey on oncology information systems using predictive analytics
2022, Research Square