Cervical cancer is a cell disease in the cervix that develops out of control in the female body. The cervix links the vagina (birth canal) with the upper section of the uterus, which can only be found in the female body. This is the second leading cause of death among women around the world. However, cervical cancer is currently one of the most preventable cancers if early detection is identified. The effect of unidentified cancer may increase the risk of death when the cell disease spreads to other parts of the female anatomy (metastasize). The Papanicolaou test is a cervical cancer screening technique used to identify potentially precancerous and cancerous cells in women’s cervix. In this paper, a few popular detection method was applying and experimented on pap smear images. A few image quality assessment (IQA) was obtained in order to determine the best of detection method. The nucleus detection will help pathologists to diagnosis in early stages of cancer. The early detection is very important stage in order to reduce the cancer incidence and mortality. The method that needs to be invented in this study is the detection method. Image detection is the process of partitioning the image into multiple regions. The detection method is object detection and recognition as well as the boundary in images. The segmented Pap Smear image is one of the detection tools with many different methods that generated different results from different issues. The solution was by analysing different existing detection methods in order to compare the dissimilar performance of existing processes. The precision of the system performance needs to be improved in order to invent a new method. As predicted from the result, the innovative construction method must be proposed and compared in order to find accurate, comprehensive measures and proper sampling procedures by the features of the selection method.
In the entire world, one of the most common cancers among women at the age of 15 until 44 years is cervical cancer disease. In most cases, it can be healed as soon as it is detected at an early phase of growth [
The Pap test method, which focuses on the nucleus cells’ morphological features, is used to detect potentially abnormal cells inside the cervix. An automated system and effective detection of the nucleus are indispensable for diagnosis [
In the thresholding process, the input’s image are convert using a grey level co-occurrence matrix (GLCM). Seven classes of cervical cancer are included in this research. After thresholding of the nucleus and extraction of its features, the feature matrix is transmitted through the neural network to categorise the image data in their relevant categories. The neural network is composed using the backpropagation technique with hidden layers [
A range of essential basic cell features (
Feature | Cervical Cancer Class | |||
---|---|---|---|---|
Normal | Degree of Dysplasia | |||
Mid | Moderate | Severe | ||
20–50 | 50+ | 50+ | 50+ | |
Dark | light | Dark | Dark | |
Light | light | Dark | Dark | |
1%–2% | 10%–20% | 20%–50% | 50% |
Thus, the studies presented to provide evidence that, according to the Algorithm of Estimation, we used the area of nucleus-to-cytoplasm ratio to detect defective cells. The k-mean clustering process can calculate the location of the nucleus. It is suitable for the detection of the biomedical image. The number of parameters (K) is typically measured by the number of similar strength regions [
In 2018, a paper from Riana et al. [
Any benefits of the proposed approach include using an automated threshold for the identification of overlapping areas. The new concept of using K-means is to detect cytoplasm and the Otsu process for overlapping regions. This research project approach also uses colour features. The overlapping area threshold is automatically performed using the Otsu method and can also be done on a cytoplasm with a diversity level. However, this researcher needs to expand the technique so that the overlapping segment areas are with several inflammatory cells [
In contrast, another study found a different technique by using K-means and Bayesian classifier on identifying the inflammatory cell and nucleus based on the categorisation of the image pixel. Three stages have been performed to identify the phase of the nucleus and inflammatory cells throughout the image. The algorithm’s procedures comprise of three phases, i.e., Image Simplification Simplifying by eliminating context, application of K-means Classification, and application of Bayesian Classification. Image Simplification simplifies an image by eliminating the background steps and transforming it into a black-and-white image. The cells and nucleus identified were previously coloured only in black backgrounds [
These results indicate the research findings from both Bayesian and K-means classification methods evaluated on sixty images in the current study. It showed that these two techniques could not identify the nucleus, inflammatory cells, and cytoplasm precisely when contrasted with the mixture of the grey-level threshold system. As can be seen from the image analysis outcomes in
In another study, a different technique was used to increase the reliability of measurement. The technique used is the Fuzzy C-means (FCM) clustering method for individual cell detection in this research study. The Fuzzy C-means clustering technique is used to segment each cell into two or three sections. The findings indicated that their detection approach provided a better set of functions for classification models. However, it is not only precision, but sensitivity is also necessary since it is a sign of fatal false negatives. Based on the results, the detection technique’s features also had excellent sensitivity [
One of the main detection problems is the overlapping of cytoplasm, which was not well discussed in past research findings. In order to resolve the overlapping problem, this study introduced a robust-shaped learning-based approach to the segment of single cells in Pap smear images to enable automated supervision of cell changes, which is a critical prerequisite for early diagnosis of cervical cancer [
Furthermore, another study on cytoplasm observation had been achieved by eliminating the surrounding portion in the image, where nothing lies, i.e., by adjusting the parameters of these pixels as if by eliminating the surrounding portion in the image or by setting the value of these pixels as 255 brings the surrounding transparent, as well as the image, remained.
In this case, two separate images of the nucleus and cytoplasm were acquired through the overlapping of Step 1 and Step 2. By doing so, researchers can identify the region overlapped between cells.
On the other side, IoU experiment results with ground truth were tested and different shape features extracted from the segmented nucleus. The classifier of the nucleus based on the form’s characteristics shall be done with the support and comparison of Random Forest Classifiers created with other classification methods and their effects on this database [
The performance analysis was evaluated using a method to determine F-measure, Sensitivity, Specificity, Accuracy, and PSNR. The method for identifying image precision is known as Image Quality Assessment (IQA). The IQA is the performance evaluation of the image, which is the essential element of the measure practice [
In this section, the PSNR block calculated the peak signal-to-noise ratio in decibels for both images. This ratio was used to calculate the performance between the original and the segmented image. The greater the PSNR, the higher the quality of the compressed or segmented image. The mean square error (MSE) and the peak signal-to-noise ratio (PSNR) were used to measure the compression quality of the image [
Based on
Based on the result shown in
where
This stacked chart shows the result of the image database from Bernsen’s method computed with the benchmark images. There are five different colours used for the segments that distinguish the categories of IQA analysis. Those are F-measure, Sensitivity, Specificity, Accuracy, and PSNR. This method showed less consistency of the sensitivity, accuracy, and F-measure in the high and low range around more than 3% to 100%. Then, the specificity was also less in consistency, but the range was higher than Otsu’s results, which were about 48% until 97% of specificity. However, the PSNR was not excellent because the lowest PNSR was about 0%.
The third method that has been analysed is Nick’s method. Nick’s binarisation acquires the adaptive threshold algorithm from the simple Niblack formula, the father of several local feature threshold techniques [
Its variance is B, and this approach moves the threshold down by adding the mean square to the variance to eliminate the background noise in the input datasets.
The chart in
Next, another analysis method is Niblack’s method. This method is known as the parent method of Nick’s method. The goal of the Niblack thresholding method is to achieve improved performance, especially for microscopic examination. It is a local detection algorithm that assimilates the threshold to the local mean and the local standard deviation over a fixed window size across each pixel location [
Since the NP is the total number of pixels in the grey image [
The chart in
Then, the fifth method analysis is Bradley’s method. This method is the short idea behind this algorithm because each pixel value is adjusted to black when its brightness is T percent smaller than the standard brightness of the pixel values in the prescribed size window; hence it is adjusted to white. This approach’s benefit is that binary images are subjectively about as beneficial as the Sauvola technique, but the measurement is two times quicker than the Sauvola approach. Sauvola’s method measures local mean as well as local variance, as Bradley’s method measures only local mean. Furthermore, since the variance can be determined with the given variance formula, as shown in
Computation of the variance re-uses the product of the measurement of the local mean (E(X))² and only measures E(X²). It takes a fair amount of period as the estimation of the local mean. Since the measurement of local mean and variance is the most time-consuming process undertaken by these two methods, Bradley’s technique is two times faster than Sauvola’s technique.
Derek Bradley stated that the Real-time adaptive thresholding method using an integral picture of the source. The methodology is an expansion of Wellner’s previous method. Besides this, our alternative is much more robust to illuminate improvements in the picture. This approach is quick and simple to execute. Bradley’s methodology is ideal for processing live video streams at a real-time frame rate, making it a powerful tool for immersive applications such as virtual reality [
This chart in
The sixth method analysed is Wolf’s technique; it solves problems in Sauvola’s technique when the text pixels and the grey-level background are close. It requires a frame or mask to measure the mean and standard deviation for implementing an adaptive threshold. The study found that the experimentally obtained results revealed that the accuracy obtained introduced by the proposed method resulted in 93.19 percent accuracy for morphological operations and 93.30 percent accuracy for the use of local Wolf thresholds [
To resolve the problems in Sauvola’s method, Wolf et al. [
When k is set to 0.5, M is the lowest grey value of the image, and R is adjusted to the total grey-value standard deviation with all local neighbourhoods (windows). Besides that, deterioration is detected when there is a sharp transformation in background grey values across the image. This is because the values of M and R are measured from the whole picture. Thus, even just a tiny, noisy spot could have a huge effect on the M and R values, eventually computing misleading image detection threshold levels [
The next method is the Feng’s adaptive threshold technique. It is appropriate since it can be qualitatively higher than the Sauvola detection algorithm. The Feng approach does, however, contain several parameters that need to be defined [
As a result,
The last method is the Sauvola method, known as the binarisation technique of Sauvola that demonstrates its robustness and efficacy when tested on low-quality databases [
R = gray-level (128), m = mean value, Δ = standard deviation and k = 0.1 [
Lastly,
In this research study, an algorithm was designed for segmenting the Pap-Smear images. All the eight methods algorithms were performed well, as shown in