IJICS JOURNAL PUBLICATION
CERVICAL CANCER DETECTION SYSTEM USING VARIOUS TEXTURAL APPROACHES
AUTHORS:
- Dr.M. Robinson Joel, Department of CSE, joelnazareth@gmail.com, SMKFIT, Chennai.
- Mr. S. Amruth, Department Of ECE, amruthsrku@gmail.com, SMKFIT, Chennai.
- Mr. J. Eyanaeswhararaj, Department Of ECE, eyanaeswh@gmail.com, SMKFIT, Chennai.
- Mr. M. Hemantha Kumar, Department Of ECE, hemanthakumar8899@gmail.com, SMKFIT, Chennai.
- Mr. J. Lokesh Department Of ECE, lokred7489@gmail.com, SMKFIT, Chennai.
ABSTRACT
Cervical cancer is one of the types of cancer which is found to be deadliest among women. Cervical cancer affects the cervix region of the uterus. The vital problem is that cancer will be identified only at the final stages and does not show any symptoms. Therefore it is necessary to give accurate treatment volume to the patient. Nowadays Image processing uses some diagnosing tools like X-ray, CT, MRI, etc. to obtain the staging of disease. This paper proposes the texture-related research works on cervical cancer detection carried out by various research communities, and authors who explore the textural feature extraction in their investigations.
Keywords: X-rays, CT, MRI, cervix, stroma, cervigram, etc.
INTRODUCTION
Cancer is the deadliest disease that is caused when cells in the human body grow old and when they die they become damaged or they do not die and form a mass of tissue that grows into a tumor. Cervical cancer occurs in women of all age groups. This cancer does not have any symptoms and it is very difficult to detect at the earlier stage of cancer development. Pap smear test is the diagnostic method proposed for women from age group 15 and above for manually screening cervical cancer. Cervical cancer is caused by the human Pamplona virus (HPV).
Different cytologists show the subjective disparity in screening the results of Pap smear tests. It provides more inconsistencies. The test output shows more false results, which makes the reliability of the screening process a question mark and the possibility of human errors become High. Thus textual approaches of image processing techniques are proposed. The texture is the predominant innate property used in the identification of certain objects or regions regardless of image types. Spectral, textural and contextural are kinds of patterns for describing the visual information meaningfully.
So it is indeed to move toward the extraction of features in a view to achieve our goal by the way of describing the medical image accurately. This section reviews the texture-related research works carried out by various research communities, and authors who explore textural feature extraction in their investigations.
Figure 1. Cervical cancer mortality statistics.
MATERIAL AND METHODS
Here we can study more about the texture classification in detail. These can be very useful in finding out the good direction towards the research. Santa & Elisa, (2017) have presented the basics of automated texture analysis in many applications of biomedicine such as the detection and grading of several types of cancer, the various diagnosis levels of autoimmune diseases, and the study of physiological processes. They have also reviewed the concepts of geometrical methods, statistical methods, local binary patterns, model-based methods, transform-based methods in the application of image segmentation, object classification, image and video compression, content-based retrieval, and 3D reconstruction and rendering. This study also investigated the latest trend i.e. deep learning architectures to learn the texture model directly from the images. It is very sensible to imagine that deep learning and more attention in the future, as its full ability in the aspect of the bio-logical textural analysis is yet to be discovered.
Figure2: visual textures with corresponding sub-patterns.
Francesco et al. (2015), have proposed the visual perception-based image features for differentiating epithelium and stroma in histological images. The following five features. i.e.coarseness, contrast, directionality, line-likeliness, and roughness were reviewed which enables functioning with a very small dimensional feature vector and allows a relevant analysis of the feature values with respect to sound visual properties. Also, three classifiers were analyzed based on Support Vector Machines (SVM), Nearest Neighbor rule (1-NN), and Naïve Bayes rule (NB). It is observed that the proposed features can appropriately distinguish epithelium from the stroma.
Figure 3. Sample images of tumor epithelium and stroma.
Jayachandran & Dhanasekaran, (2014) have presented a brain tumor classification method by incorporating structural analysis on both timorous and normal tissues. This framework involves the preprocessing, segmentation, feature extraction and classification steps to be followed in detecting tumors. As a preprocessing step, the anisotropic filter is applied to eradicate the noise which improves the image quality for the skull-stripping process. Some of the features are captured from the intensity and some are from the modified multi-texton structure descriptor. The hybrid kernel is formulated to train the SVM classifier to carry out automatic categorization of tumors in magnetic resonance imaging (MRI) images.
Figure 4: sample MRI Image Dataset, (a) Normal image, (b) tumor image.
Sun, Park, et al (2011) have developed a domain-specific computerized image analysis for the identification of pre-cancerous and cancerous lesions of the cervix. Here, features are extracted in a probabilistic manner using conditional random field and also introduced a window-based assessment scheme for 2D image analysis which solves the missed alignment problem. Various tissue types of image regions are obtained for the extraction of domain-oriented anatomical features. Diagnostic relationship between tissue types is also obtained using a conditional random field.
Figure 5: (a) Cross polarized white light image; and (b) histopathology slide of a patient's cervix
Hayit et al. (2009) have proposed a multi-stage scheme for segmenting and labeling of the anatomical region inside the cervigrams and also features of cervix region are extracted thereby detecting cervix boundary. Besides that, specular reflection has been destroyed in the pre-processing step, and entry of the endo-cervical canal is also detected. Cervigram analysis includes a two-step process in which undesirable image regions are ignored by an ROI detection method and as a second step, specular reflection has been detected and removed. Geometric measure of local concavity method is incorporated for detection of the os (neck of the uterus).
Figure. 6. An example cervigram: the cervix boundary, the os, and SR artifacts are marked.
Figure 7. Typical vascular patterns encountered in cervical lesions. (a) network capillaries in the original squamous epithelium; (b) hairpin capillaries in the original squamous epithelium; (c) and
(d) punctation vessels in dysplasia and carcinoma in situ; and (e) and (f) mosaic vessels as seen in dysplasia and carcinoma in situ
Figure 8. Cervical cancer incidence rate and mortality rate in various regions of the world.
Table 2.1 provides the detailed report of the existing cervical cancer detection frameworks which are emphasized textural features. On proper monitoring of the merits and demerits of these papers, some of the papers do not consider textural features, and only basic methods were applied for extracting textural features like first-order statistics features and simple co-occurrence patterns. These features do not exhibit efficient feature representation for increasing the performance of the cancer detection systems in terms of precision and recall.
Table 2.1 Textural feature-based Cancer detection system
CONCLUSION
Cervical cancer is screened by the manual method (pap smear test) does not give accurate results in classifying the tumoured cells and normal cells located at the cervix of the uterus. In this paper, a study and analysis are done for detecting cervical cancer automatically using various textural approaches. The study was made to investigate the features derived from image processing of MR images of patients that could be used in predicting the disease stage. Thus, textural approaches outperformed transform and statistical features in staging prediction. This idea which has been proposed by many journalists and researchers is taken as a survey in this paper. From this, we come to a conclusion that the textural approaches will become the best method of diagnosing the earlier stages of cervical cancer. It is better to predict the treatment volume will help the radiologist for better treatment planning according to the staging.
REFERENCES
[1] Mining textual knowledge in biological images: Applications, methods, and trends: San- ta Di Cataldo, Elisa Ficarra.
[2] A, Jayachandran & Raghavan, Dhanaseka- ran. (2014). Brain Tumor Severity Analysis Using Modified Multi-Texton Histogram and Hybrid Kernel SVM. International Journal of Imaging Systems and Technology. 24. 10.1002/ima.22081.
[3] Park, Sun & Sargent, Dustin & Lieberman, Richard & Gustafsson, Ulf. (2011). Domain-Specific Image Analysis for Cervical Neoplasia Detection Based on Conditional Random Fields. IEEE transactions on medical imaging. 30. 867- 78. 10.1109/TMI.2011.2106796.
[4] Hayit & Gordon, Jose & Antani, Sameer & Long, L. (2009). Automatic Detection of Anatomical Landmarks in Uterine Cervix Images. IEEE transactions on medical imaging. 28. 454- 68. 10.1109/TMI.2008.2007823.
[5] Ji, Qiang & Engel, John & Craine, Eric. (2000). Texture Analysis for Classification of Cervix Lesions. IEEE transactions on medical imaging. 19. 1144-9. 10.1109/42.896790.
[6] Bianconi, Francesco & Alvarez-Larrán, Al- berto & Fernández, Antonio. (2015). Discrimi- nation between tumour epithelium and stroma via perception-based features. Neurocomputing. 154. 119–126. 10.1016/j.neucom.2014.
0 comments:
Post a Comment