https://orcid.org/0000-0001-7961-0087
Figures
Abstract
So far, the number of patients who die from cancer is quite high. Continuation of early detection research is important to reduce the number of deaths due to cancer. At the time of the literature review, images of the same patients taken from Scanning Electron Microscope (SEM) and Atomic Force Microscope (AFM) for early diagnosis of cervix cancer have not been addressed to date. This article, Photodiagnosis and Photodynamics with SEM and AFM images are valuable in recognizing cervical cancer and starting treatment early. Simultaneous examination of the, Photodiagnosis and Photodynamics with SEM and AFM cervix images of patients will provide us with a far more powerful solution than a one-way solution. Daubechies (db2, db3, db4, and db5), Coiflet (coif5, coif4, coif3, and coif2), Symlet (sym5, sym4, sym3, and sym2), and Biorthogonal (bior1.3, bior2.8, bior1.5, and bior3.3) 16 discrete wavelet transformation families (DWTF) have been applied to AFM and SEM images. One approximate and three detail coefficients have been obtained for each one AFM and SEM cervix images. Homogeneity, contrast, angular second moment, entropy, mean, standard deviation, correlation, cluster prominence, dissimilarity, and cluster shade values have been calculated for each of these one approximate and three detail coefficients. The classification rate found by the averages of the results obtained from the DWTF_JSD, DWTF_HD and DWTF_TD algorithms for AFM and SEM cervix images are 98.29% and 97.10%, respectively. According to these results, it has been determined that SEM images have lower classification rate than AFM images. It has been also observed that the surface roughness of the mAFM images was larger than nAFM and bAFM images. But, it was observed that the volume of particles of the mAFM images has been smaller than nAFM and bAFM images.
Citation: AYTAÇ S, ÖZBEY G (2025) Photodiagnosis and photodynamic recognition of cervical cancer with SEM and AFM images. PLoS ONE 20(2): e0316544. https://doi.org/10.1371/journal.pone.0316544
Editor: Mallikarjuna Reddy Kesama, Purdue University, UNITED STATES OF AMERICA
Received: August 31, 2024; Accepted: December 12, 2024; Published: February 6, 2025
Copyright: © 2025 AYTAÇ, ÖZBEY. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: We received the data from the hospital with non-invasive ethics committee permission. It is not legal to share the data we receive regarding ethical permission. Required Ethics committee permission documents are available in additional files named "other". Other "PLOSOne_Human_Subjects_Research_Checklist.docx" Other "Aug 22, 2024 Doc (1).pdf" Other "etik kurul onay1.pdf" Other "etik ingilizce.doc" The data is the information of cervical cancer patients. Therefore open access cannot be allowed. Since the data cannot be shared publicly, it can be provided by the corresponding author of the study upon reasonable requests. If you have any questions or requests, you can contact me at my email "ozbeyg@hotmail.com". Cervic cancer data were recorded on glass coverslips with the permission of Fırat University's non-invasive ethics committee and with the approval of Prof. Dr. İbrahim ÖZERCAN, the head of the Fırat University pathology department. Data access restriction has been made because the data received contains sensitive data such as identity information and examination dates of cancer patients. The e-mail address of pathologist Prof. Dr. İbrahim ÖZERCAN, head of the department from which data was obtained, is ozercanih@firat.edu.tr. In our article, Prof. Dr. İbrahim Özercan, head of the department of pathology, was thanked for his data contribution. email adress: ozercanih@firat.edu.tr The e-mail address of the corresponding author is ozbeyg@hotmail.com. Dr.İbrahim ÖZERCAN's e-mail address is ozercanih@firat.edu.tr. Interested researchers may reach out to Dr.İbrahim ÖZERCAN for data request as well as the corresponding author.
Funding: The author(s) received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
1. Introduction
Scanning Electron Microscopy (SEM) [1] was invented for the first time in 1935 by a researcher, Knoll, to study the surfaces of solid bodies [2]. Examination of the surface characteristics of cells with SEM reveals the findings that can not be obtained with light microscopy and transmission electron microscopy [2, 3]. Obtaining the image with a SEM microscope is the result of evaluation of the incoming electron beams after reflection from the sample surface [2, 4]. AFM is a type of scanning tip microscope. The AFM developed by Heinrich Rohrer and Gerd Binnig about 20 years ago, works by detecting a sample surface of a mechanical tip. The device is a very high resolution scanning force microscope, but it also involves investigating the mechanical properties of the surfaces, even of nanometer scale samples. The most important advantage is that the samples can be viewed directly in any environment without special preparation. Additionally, nanobiomechanical quantities such as surface roughness and particle volume can be measured with AFM. Surface roughness, often abbreviated as roughness, is a component of surface texture. It is measured by the deviations in the direction of the normal vector from the ideal form of a real surface. If these deviations are large, the surface is rough; If they are small, the surface is smooth. Its unit is (nanometer (nm)). Also, unit of particle volume is (micrometer (μm)). A micrometer is one thousandth of a millimeter and one millionth of a meter [2, 5, 6].
Nowadays, it has become a very important issue the extraction of the effective features for texture-based and content-based images [7]. Nowadays, the analysis of tissue-type images is used in many fields such as medical imaging, industrial research, document segmentation and classification, recognition and classification [8]. Gaussian Markov Random Fields (GMRF) which be probabilistic models and autoregulation methods has been investigated for the analysis of another study [8, 9]. Various methods have been developed for the analysis of tissue type images. Statistical methods are one of these [10, 11]. Analysis of texture-type images have recently shifted to multi-resolution time-frequency methods such as Gabor filters [12] and wavelet decomposition [13–15]. These methods provide very good possibilities for analyzing and classifying tissue type images [13–15]. In order to fully analyze, categorize and recognize the tissue type images, the distinguishing features to be extracted from these images must be sufficient. We can classify the studies that used old feature extraction methods used for analysis of texture-like images to be as follows:
Studies using statistical features of first and second order of tissue type images [16, 17]. Studies using Gibbs Random Fields and GMRF are in references [18–20]. Studies recommending of local linear transformations are in references [21, 22]. These classical methods must largely show the same characteristics of different regions of tissue type sections. However, some of the restrictions given above still cannot ensure this. However, the biggest problem encountered in texture analysis by the above-mentioned classical methods is that the same or similar characteristics can not be obtained from each of the randomly selected images of the same size in the image. Therefore, in this article wavelet transformation method, which has become one of the most popular methods in tissue type image analysis, is used. Because wavelet transform offers the convenience of expressing that these groups of pixels are the same pieces of image by introducing robust features that can minimize the characteristic difference between different regions of the same size selected randomly [23]. Furthermore, the wavelet transform improves the robust recognition performance of the images in different metrics with the multi-resolution feature it brings. One of the biggest factors is that the high and low bands used for DWT can remain the identical for a fraction of different consecutive scales [23]. The main objectives of this study can be summarized as follows:
- To demonstrate the performance of the DWTF method which combined with Homogeneity, contrast, angular second moment, entropy, mean, standard deviation, correlation, cluster prominence, dissimilarity, and cluster shade properties of cervical AFM and SEM cells.
- To identify cervical AFM and SEM cells using wavelet transform based JSD, HD, and TD algorithms in the analysis of cervical images.
In this article, normal SEM cells are nSEM. Benign SEM cells are bSEM. Malign SEM cells are mSEM. Also, normal AFM cells are nAFM. Benign AFM cells are bAFM., and malign AFM cells are mAFM. SEM cells are acquired from Sciece Faculty SEM Laboratory in the Fırat University. Besides, AFM cells are acquired from Nanotechnology Laboratory in the Fırat University. In this article a total of 45 images of each the cervix cells (nSEM, bSEM, mSEM nAFM, bAFM, and mAFM) type were used in 512 x 512 sizes. Every of these cervical cells was described into regions, which consists 4 pieces of nonintersecting 256x256 cervical cells. Therefore, 3x180 quantities were occurred from cervical cells which have 3x4x45 quantities. So, total 540 SEM cells have been taken as 3x4x45 (nSEM, bSEM, and mSEM). Also, total 540 AFM cells have been taken as 3x4x45 (nAFM, bAFM, and mAFM). The 270 pieces of these 540 and the remaining 270 pieces of them for both SEM cells and for AFM cells were used for training and testing steps of suggested algorithm in this study separately. For this purpose, 16 DWTFVs are applied to each of the 256 × 256 image regions obtained for each of the 45 images. Then, one approximate and three detail coefficients has been calculated with DWT method according to level 3 for each of the 256 × 256 size images. Homogeneity, contrast, angular second moment, entropy, mean, standard deviation, correlation, cluster prominence, dissimilarity, and cluster shade features have been calculated for each of the these approximate and detail coefficients. Thus, DWTFVs in the 4x10 size have been obtained for each 256x256 image. Homogeneity, contrast, angular second moment, entropy, mean, standard deviation, correlation, cluster prominence, dissimilarity, and cluster shade feature values were given for training to the DWTF_JSD, DWTF_HD and DWTF_TD algorithms.
This article can be summarized as follows: In Chapter 1 has been discusses from previous work. In Section 2 has been given general information’s in about the DWTF, DWTF_JSD, DWTF_HD and DWTF_TD algorithms. The recommended system steps in section 3 are described. Experimental study results of categorization of cervical images in section 4 are given. The results obtained in Section 5 are discussed.
2. Material and methods
The study adhered to the Declaration of Helsinki and was approved by the local ethics review board. In this study, we used real medical images in order to evaluate the cervical cancer diagnosis. We sought Non-Interventional Clinical Researches Ethics Board in Firat University in order to obtain patient records coming from Elazig and its neighbor cities in Turkey. The medical image usage was approved with decision number of 07–05 by the related board. After this, we collected the patients’ medical records belonging to Firat University Hospital. Patients’ medical records were taken suitably to the ethics rules of the hospital. Medical tissue samples are determined from Firat University Medicine Faculty Pathology Laboratory. Our study is a retrospective study. Ethical approval was obtained before starting the study. It was conducted in accordance with the Declaration of Helsinki. Our relevant ethics committee approval has been uploaded to the other files sections. Our study started on 01/05/2016 and ended on 01/05/2021. Our study is a retrospective study. The data used in the study were obtained from medical archive records after ethical approval. Our ethics committee therefore did not require a consent form and waived the need for consent. Therefore, consent was not obtained. No patient data under the age of 18 was used in the medical data used in the study. The medical records required for our study were accessed between 01/05/2016 and 01/05/2021. All data were obtained from archived medical records after obtaining relevant permissions and approvals. No extra intervention was given to the participants.
2.1. Discrete Wavelet Transform Families (DWTF)
Biorthogonal (bior), Symlet (sym), Coiflet (coif), and Daubechies (db) discrete wavelet transformation families have been applied to AFM and SEM images. The db2, db3, db4, and db5 which be types of the db wavelet family have been applied to the AFM and SEM images. The coif2, coif3, coif4, and coif5 which be types of the Coif wavelet family have been applied to the AFM and SEM images. The sym2, sym3, sym4, and sym5 which be types of the sym wavelet family have been applied to the AFM and SEM images. The bior1.3, bior2.8, bior1.5, and bior3.3 which be types of the bior wavelet family have been applied to the AFM and SEM images.
2.1.1. Daubechies wavelet.
Daubechies is the orthogonal property of the wavelike family and has asymmetrical properties that cause large phase distortions. Therefore, the Daubechies wavelet family should not be used in applications where information must be kept [24].
2.1.2. Coiflet wavelet.
Coiflet wavelet family is steep, and shows a close symmetry feature. This close symmetry feature ensures that the Coiflet ripple is of near linear phase [24].
2.1.3. Symlet wavelet.
Symlet wavelet is vertical, and is in feature of close symmetric. This feature means low information deterioration. At Nth grade, a Symlet wave has N number of collapse torques for a support width of 2N-1. Symlet wavelets are very similar to Daubechies wavelets, except for its very symmetrical feature [24].
2.1.4. Biorthogonal wavelet.
Biorthogonal wavelet families are upright and symmetrical. Such main wavelets are done by spline method [24].
Wavelet Transform is a different transformation method developed to overcome the shortcomings of non-stationary signals of the Fourier transform [13]. This analysis method is less sensitive to noise and can be easily applied to non-stationary signals. Therefore, the interest of those engaged in signal processing has shifted from the frequency-based Fourier Transform to the scale-based Wavelet Transform. The wavelet statement was put forward by Alfred Haar in 1909 [25]. Over time, Jean Marlet and Y. Meyer and their colleagues developed this method and in 1988, Stephane Mallat made in significant contributions. Later, international researchers such as Ingrid Daubechies and Ronald Coifman developed the method and brought it to the present. Wavelet transform has recently become a widely used method for multi-resolution analysis of signatures and images, such as radar target recognition, communication, texture image classification, and many other inland feature extraction [26].
The biggest advantage of wavelet transforms is that these transformations have a variable window size that window-size is wide for low frequencies and narrow for high frequencies. This ensures an optimum time-frequency resolution over all frequency ranges [13–15]. Discrete wavelet transform is a widely used method for feature extraction in recent years, such as signal processing, biomedical signal processing, and communication [14, 15]. The Discrete Wavelet Transform (DWT) is the sum of the signal that is scaled over time and multiplied by the offset values of the wavelet over time. The DWT function can be used at the desired level for the image or signal [23]. The DWT lower band separation method allows to obtain the discriminative characteristics of the signal by analyzing the component of the high frequency in a narrow range and the component of the low frequency in a wider band interval. The mathematical expression of the sub band separation up to the fourth level using DWT is shown in Eq (1) [27].
Fig 1 shows the process of separating the mark to one level lower bands.
The g [n] in Fig 1 represents the high pass filter and the h [n] low pass filter. In the decomposition process for these sub-bands, h [n] expresses approximate coefficients for the operation, while g [n] expresses the detail coefficients for the operation [27].
Signaling is first subjected to filtering by line basis. As a result of this filtering, two groups of data are acquired without low and high pass filtering. Sample reduction is performed on the acquired data. These two subgroups are then again passed through low and high-pass filters according to the column principle. Two subgroups of data are acquired from these two subgroups. As a result, four sub data groups are acquired: horizontal, vertical and diagonal [14, 15, 27]. The approach data can be resumed in a similar way and DWT can be resumed. In Fig 1, CA approach, CD1 horizontal, CD2 vertical, CD3 diagonal is coefficient values.
2.2. Classification
2.2.1. Jensen Shannon Divergence (JSD)
JSD is one of the methods used to measure the distance between two probability distributions. In the literature, "information radius" or total divergence to average is also passed. By definition, the Kullback-Leibler Divergence can be thought of as a symmetrical state. That is, the distances from t to u for t and u, which are two probability distributions are not equal to every other at the distance of Kullback Leibler. The DKL (t || u)) distance which be from t to u with (DKL (t || u)) distance are not equal to every other [11]. However, JSD’s distance solves this problem and defines a symmetric distance function. The basic definition is as follows in Eq (2) [23, 28]:
(2)
2.2.2. Hellinger Divergence (HD).
HD is the distance function in the between t and u. The HD distance function [29] is shown in Eq (3).
2.2.3. Triangle Divergence (TD).
TD is distance function in the between t and u. The TD distance function [23, 28] is shown in Eq (4).
(4)
(5)
(6)
where WJSDav,HDav,TDav (p) are average weight calculation, WJSD,HD,TD (p) are weight of the extracted p attribute for JSD, HD and TD, respectively. Total of cell numbers is K, z is standardization fixed, apr is r-th value in p-th feature, the goal attribute is k, P(t) is probabilistic value of the cells, nSEM+nAFM, bSEM+bAFM and mSEM+mAFM.
2.2.4. Classification with DWTF-WJSD, DWT-WHD, and DWTF-WTD algorithms.
These weights are written in the following in Eq (7), after calculating the weight for the cells [23, 30].
(7)
where the goal attribute is c, dJSD is test value of the JSD., dHD is test value of the HD, dTD is test value of the TD.
2.2.5. Average of the DWTF-WJSD, DWT-WHD, and DWTF-WTD algorithms.
Average ClassificationSEM and AFM is average of the DWTF-WJSD, DWT-WHD, and DWTF-WTD Algorithms in Eq (8).
(8)
where (dJSD)SEM is test value of the JSD for SEM images., (dHD)SEM is test value of the HD for SEM images, (dTD)SEM is test value of the TD for SEM, (dJSD)AFM is test value of the JSD for AFM images., (dHD)AFM is test value of the HD for AFM images, (dTD)AFM is test value of the TD for AFM. Average ClassificationSEM and AFM is average of the DWTF-WJSD, DWT-WHD, and DWTF-WTD Algorithms for SEM and AFM images.
3. Results
SEM cells in this article are nSEM, bSEM, and mSEM images. Also, AFM cells are nAFM, bAFM, and mAFM images. SEM cells are acquired from Sciece Faculty SEM Laboratuary in the Fırat University. Besides, AFM cells are acquired from Nanotechnology Laboratuary in the Fırat University. This article can be summarized as follows: Firstly, 45 SEM and AFM cervix images in the 512 × 512 size have been used for each total image type (nSEM, bSEM, mSEM nAFM, bAFM, and mAFM) obtained from Fırat University. The size of the cervical SEM and AFM images obtained from each of these images is 256 × 256. Therefore, 4x45 images were obtained from each type of image. So, total 540 SEM cells have been taken as 3x4x45 (nSEM, bSEM, and mSEM). Also, total 540 AFM cells have been taken as 3x4x45 (nAFM, bAFM, and mAFM).
The 2x270 images which taken from these 256 × 256 image regions obtained for each SEM and AFM (2 type) servix image were used in the training phase of DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms. The remaining 2x270 images are used in the testing phase of these algorithms. Each of these 256 × 256 image parts obtained for each of the 45 cervix SEM and AFM images was applied Biorthogonal (bior), Symlet (sym), Coiflet (coif), and Daubechies (db) wavelet families. For this, the coif2, coif3, coif4, and coif5 which be types of the Coif wavelet family have been used for the SEM and AFM images. The sym2, sym3, sym4, and sym5 which be types of the sym wavelet family have been used for the SEM and AFM images. The bior1.3, bior2.8, bior1.5, and bior3.3 which be types of the bior wavelet family have been used for the SEM and AFM images. And, the DWT coefficients of these images are calculated according to Level 3.
Then, one approximate and three detail coefficients has been calculated with DWT method according to level 3 for each of the 256 × 256 size images. Homogeneity, contrast, angular second moment, entropy, mean, standard deviation, correlation, cluster prominence, dissimilarity, and cluster shade features have been calculated for each of the these approximate and detail coefficients. Thus, DWTFVs in the 4x10 size have been obtained for each of the 540 AFM and 540 SEM cervix images. The 270 pieces of these 540 and the remaining 270 pieces of them for both SEM cells and for AFM cells were used for training and testing steps of suggested algorithm in this study separately. The structure of DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms is shown in Fig 2. This structure is used for classification of cervical SEM and AFM images.
Accuracy of classification results using DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms in the Fig 2 has been tested. For this aim, 45 SEM and 45 AFM cervix images in the 512 × 512 size in this study have been used for type of the each image (nSEM, bSEM, mSEM nAFM, bAFM, and mAFM). The size of the SEM and AFM cervix images obtained from each of these images is 256 × 256. Therefore, 4x45 images were obtained from each type of image. So, total 540 SEM cells have been taken as 3x4x45 (nSEM, bSEM, and mSEM). Also, total 540 AFM cells have been taken as 3x4x45 (nAFM, bAFM, and mAFM). The 2x270 images which taken from these 256 × 256 image regions obtained for each SEM and AFM (2 type) servix image were used in the training phase of DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms. The remaining 2x270 images are used in the testing phase of these algorithms.
4. Discussions
SEM cells in this article are nSEM, bSEM, and mSEM images. Also, AFM cells are nAFM, bAFM, and mAFM images. SEM cells are acquired from Sciece Faculty SEM Laboratory in the Fırat University. Besides, AFM cells are acquired from Nanotechnology Laboratory in the Fırat University. nSEM is seen in Fig 3(a). bSEM is seen in Fig 3(b). mSEM is seen in Fig 3(c).
(a) nSEM (b) bSEM (c) mSEM.
Fig 4(a) shows the nAFM cervix cell images. Fig 4(a) is an image the two dimensional. Three dimensional image of nAFM is seen in Fig 4(b). Particle volume and surface roughness values of the nAFM image are seen in Fig 4(c). Values the of surface roughness and particle volume for nAFM images are measured as 50.23 nm and 4.375 μm.
(a) Two-Dimensional nAFM cervix image (b) Three-dimensional nAFM cervix image (c) Particle volume and surface roughness nAFM cervix image.
Fig 5(a) shows the bAFM cervix cell images. Fig 5(a) is an image the two dimensional. Three dimensional image of bAFM is seen in Fig 5(b). Particle volume and surface roughness values of the bAFM image are seen in Fig 5(c). Values the of surface roughness and particle volume for bAFM images are measured as 244.654 nm and 4.271 μm.
(a) Two-Dimensional bAFM cervix image (b) Three-dimensional bAFM cervix image (c) Particle volume and surface roughness bAFM cervix image.
Fig 6(a) shows the mAFM cervix cell images. Fig 6(a) is an image the two dimensional. Three dimensional image of mAFM is seen in Fig 6(b). Particle volume and surface roughness values of the nAFM image are seen in Fig 6(c). Values the of surface roughness and particle volume for mAFM images are measured as 456.145 nm and 3.125 μm.
(a) Two-Dimensional mAFM cervix image (b) Three-dimensional mAFM cervix image (c) Particle volume and surface roughness mAFM cervix image.
Fig 7 represents the mSEM cervix image at level 3 which used sym3 DWTFV.
Fig 8 represents the mAFM cervix image at level 3 which used sym2 DWTFV.
45 SEM and 45 AFM cervix images in the 512 × 512 size in this study have been used for type of the each image (nSEM, bSEM, mSEM nAFM, bAFM, and mAFM) obtained from Fırat University. The size of the cervical SEM and AFM images obtained from each of these images is 256 × 256. Therefore, 4x45 images were obtained from each type of image. So, total 540 SEM cells have been taken as 3x4x45 (nSEM, bSEM, and mSEM). Also, total 540 AFM cells have been taken as 3x4x45 (nAFM, bAFM, and mAFM). The 2x270 images which taken from these 256 × 256 image regions obtained for each SEM and AFM (2 type) servix image were used in the training phase of DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms. The remaining 2x270 images are used in the testing phase of these algorithms. The performance of this Cervical SEM and AFM image classification obtained using the DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms is given in Tables 1 and 2. Average the classification rate for DWTFVs with DWTF-WJSD algorithm for AFM images in Table 1 is obtained as 98.35%. average the classification rate for DWTFVs with DWTF-WHD algorithm for AFM images is obtained as 98.09%. Average the classification rate for DWTFVs with DWTF-WTD algorithm for AFM images is obtained as 98.45%. Average of these averages founded from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms are obtained as 98.29%.
Average the classification rate for DWTFVs with DWTF-WJSD algorithm for SEM images in Table 2 is obtained as 97.15%. average the classification rate for DWTFVs with DWTF-WHD algorithm for SEM images is obtained as 97.08%. Average the classification rate for DWTFVs with DWTF-WTD algorithm for SEM images is obtained as 97.09%. Average of these averages founded from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms are obtained as 97.10%
5. Conclusion
To date, there are different studies in the field of machine learning in different types of cancer [28–30]. However, to date, many studies have been carried out in the field of biomedicine and different types of cancer [31–36]. However, to date, there is no other study using AFM and SEM images together for early cervical cancer diagnosis. SEM cells in this article are nSEM, bSEM, and mSEM images. Also, AFM cells are nAFM, bAFM, and mAFM images. SEM cells are acquired from Sciece Faculty SEM Laboratory in the Fırat University. Besides, AFM cells are acquired from Nanotechnology Laboratory in the Fırat University. 270 AFM and 270 SEM images were used for training purposes. 270 AFM and 270 SEM images were used for testing purposes. Classification rates of the nAFM, bAFM, and mAFM cervix images are shown in Table 1.
Similarly, Classification rates of the nSEM, bSEM, and mSEM cervix images are shown in Table 2.
These AFM classification rates found in Fig 9 are obtained from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms.
These SEM classification rates found in Fig 10 are obtained from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms.
Average the classification rate for DWTFVs with DWTF-WJSD algorithm is obtained as 98.35% for AFM images in Fig 11.
Average the classification rate for DWTFVs with DWTF-WHD algorithm is obtained as 98.09% for AFM images. Average the classification rate for DWTFVs with DWTF-WTD algorithm is obtained as 98.45% for AFM images. Average of these averages founded from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms are obtained as 98.29%. Average the classification rate for DWTFVs with DWTF-WJSD algorithm is obtained as 97.15% for SEM images in Fig 11. Average the classification rate for DWTFVs with DWTF-WHD algorithm is obtained as 97.08% for SEM images. Average the classification rate for DWTFVs is obtained as 97.09% with DWTF-WTD algorithm for SEM images. Average of these averages founded from DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms are obtained as 97.10%.
4.375 μm, 4.271 μm and 3.125 μm seen in Fig 12 are the particle volume values of the nAFM, bAFM, and mAFM images, respectively. Also, mAFM particle volume value is smaller than nAFM and bAFM particle volume value.
50.23 nm, 244.654 nm and 456.145 nm seen in Fig 13 are the surface roughness values of the nAFM, bAFM, and mAFM images, respectively. Also, mAFM surface roughness value is smaller than nAFM and bAFM surface roughness value.
The fact that it has a fast and expensive structure is one of the advantages of DWTF-WJSD, DWTF-WHD, and DWTF-WTD algorithms. This system can be used to classify images of ultrasonographic, mammographic, endoscopic and different types of cancer in subsequent studies.
Acknowledgments
The authors state that they thank to Prof. Dr. İbrahim Hanifi ÖZERCAN, head of Pathology Department of Firat University, Faculty of Medicine. Additionally, the authors state that they thank to Prof. Dr. Fahrettin YAKUPHANOĞLU, Fırat University Physics Department faculty member for AFM and SEM images.
References
- 1. MCMULLAN D. An improved scanning electron microscope for opaque specimens. Proceedings of the IEE-Part II: Power engineering, 1953, 100.75: 245–256.
- 2.
GERÇEKER, Funda. Bakteri-Epitel Hücre İlişkisinin Işık ve Scanning (Taramalı) Elektron Mikroskobu İle İncelenmesi. 2013.
- 3. WILLIAMS Anthony E., et al. The surface ultrastructure of normal and metaplastic cervical epithelia and of carcinoma in situ. Cancer Research, 1973, 33.3: 504–513. pmid:4689899
- 4. AKÇAYÖZ Deniz, et al. Elektron mikroskobun tipta kullanim alanlari. 2010.
- 5. BINNIG Gerd; QUATE Calvin F.; GERBER Ch. Atomic force microscope. Physical review letters, 1986, 56.9: 930. pmid:10033323
- 6. KASAS S., et al. Biological applications of the AFM: from single molecules to organs. International journal of imaging systems and technology, 1997, 8.2: 151–161.Lin, C. H., Chen, R. T., & Chan, Y. K. (2009). A smart content-based image retrieval system based on color and texture feature. Image and Vision Computing, 27(6), 658–665.
- 7. SENGUR Abdulkadir; GUO Yanhui. Color texture image segmentation based on neutrosophic set and wavelet transformation. Computer Vision and Image Understanding, 2011, 115.8: 1134–1144.
- 8. SENGUR Abdülkadir. AN UNSUPERVISED IMAGE SEGMENTATION USING POISSON MARKOV RANDOM FIELDS. Technological Applied Sciences, 2007, 2.4: 305–321.
- 9. KONUKOGLU Ender, et al. Multivariate statistical analysis of diffusion imaging parameters using partial least squares: application to white matter variations in Alzheimer’s disease. Neuroimage, 2016, 134: 573–586. pmid:27103138
- 10.
KORKMAZ Sevcan Aytaç; EREN Haluk. Cancer detection in mammograms estimating feature weights via Kullback-Leibler measure. In: 2013 6th international congress on image and signal processing (CISP). IEEE, 2013. p. 1035–1040.
- 11. KONG Wai Kin; ZHANG David; LI Wenxin. Palmprint feature extraction using 2-D Gabor filters. Pattern recognition, 2003, 36.10: 2339–2347.
- 12. ŞEN Baha, et al. A comparative study on classification of sleep stage based on EEG signals using feature selection and classification algorithms. Journal of medical systems, 2014, 38: 1–21.
- 13. SENGUR Abdulkadir; TURKOGLU Ibrahim; INCE M. Cevdet. Wavelet packet neural networks for texture classification. Expert systems with applications, 2007, 32.2: 527–533.Sengur, A., Turkoglu, I., & Ince, M. C. (2008). Wavelet oscillator neural networks for texture segmentation. Neural Network World,18(4), 275.
- 14. CUADRA Meritxell Bach, et al. Comparison and validation of tissue modelization and statistical classification methods in T1-weighted MR brain images. IEEE transactions on medical imaging, 2005, 24.12: 1548–1565. pmid:16350916
- 15.
DOYLE, Scott, et al. Automated grading of prostate cancer using architectural and textural image features. In: 2007 4th IEEE International Symposium on Biomedical Imaging: From Nano to Macro. IEEE, 2007. p. 1284–1287.
- 16.
ONURAL, Levent. Gibbs random fields and Markov random fields with constraints. arXiv preprint arXiv:1603.01481, 2016.
- 17.
STOEHR Julien; FRIEL Nial. Calibration of conditional composite likelihood for Bayesian inference on Gibbs random fields. In: Artificial Intelligence and Statistics. PMLR, 2015. p. 921–929.
- 18. GHAMISI Pedram; BENEDIKTSSON Jon Atli; ULFARSSON Magnus Orn. Spectral–spatial classification of hyperspectral images based on hidden Markov random fields. IEEE Transactions on Geoscience and Remote Sensing, 2013, 52.5: 2565–2574.
- 19. HARANDI Mehrtash T.; NILI AHMADABADI Majid; ARAABI Babak N. Optimal local basis: A reinforcement learning approach for face recognition. International Journal of Computer Vision, 2009, 81: 191–204.
- 20. KIM Tae-Kyun; KITTLER Josef. Locally linear discriminant analysis for multimodally distributed classes for face recognition with a single model image. IEEE transactions on pattern analysis and machine intelligence, 2005, 27.3: 318–327. pmid:15747788
- 21. KORKMAZ S. Aytac; POYRAZ Mustafa. A new method based for diagnosis of breast cancer cells from microscopic images: DWEE—JHT. Journal of medical systems, 2014, 38: 1–9.
- 22. ALI Mohammed Nabih; EL-DAHSHAN EL-Sayed A.; YAHIA Ashraf H. Denoising of heart sound signals using discrete wavelet transform. Circuits, Systems, and Signal Processing, 2017, 36: 4482–4497.
- 23.
TANYEL, Emre Doruk. Formation evaluation using wavelet analysis on logs of the Chinji and Nagri Formations, northern Pakistan. 2006. PhD Thesis. Texas A&M University.
- 24. DAUBECHIES Ingrid, et al. Wavelet transforms and orthonormal wavelet bases. Different perspectives on wavelets, 1993, 47: 1–33.
- 25.
KARHAN, Zehra; ERGEN, Burhan. Content based medical image classification using discrete wavelet and cosine transforms. In: 2015 23nd Signal Processing and Communications Applications Conference (SIU). IEEE, 2015. p. 1445–1448.
- 26.
LU, Chao, et al. Precise segmentation of multiple organs in CT volumes using learning-based approach and information theory. In: Medical Image Computing and Computer-Assisted Intervention–MICCAI 2012: 15th International Conference, Nice, France, October 1–5, 2012, Proceedings, Part II 15. Springer Berlin Heidelberg, 2012. p. 462–469.
- 27. LEE Chang-Hwan. A Hellinger-based discretization method for numeric attributes in classification learning. Knowledge-Based Systems, 2007, 20.4: 419–4
- 28. KORKMAZ Sevcan Aytac; KORKMAZ Mehmet Fatih. A new method based cancer detection in mammogram textures by finding feature weights and using Kullback–Leibler measure with kernel estimation. Optik, 2015, 126.20: 2576–2583.
- 29. KORKMAZ Sevcan Aytac; KORKMAZ Mehmet Fatih; POYRAZ Mustafa. Diagnosis of breast cancer in light microscopic and mammographic images textures using relative entropy via kernel estimation. Medical & biological engineering & computing, 2016, 54: 561–573. pmid:26345243
- 30. KORKMAZ Sevcan Aytac, et al. Diagnosis of breast cancer nano-biomechanics images taken from atomic force microscope. Journal of Nanoelectronics and Optoelectronics, 2016, 11.4: 551–559.
- 31. AMINA Musarat, et al. Biogenic green synthesis of MgO nanoparticles using Saussurea costus biomasses for a comprehensive detection of their antimicrobial, cytotoxicity against MCF-7 breast cancer cells and photocatalysis potentials. PLoS One, 2020, 15.8: e0237567. pmid:32797097
- 32. TAGHIZADEH Mohammad Sadegh, et al. Experimental, molecular docking and molecular dynamic studies of natural products targeting overexpressed receptors in breast cancer. PloS one, 2022, 17.5: e0267961. pmid:35536789
- 33. ANGUIANO María, et al. The use of mixed collagen-Matrigel matrices of increasing complexity recapitulates the biphasic role of cell adhesion in cancer cell migration: ECM sensing, remodeling and forces at the leading edge of cancer invasion. PloS one, 2020, 15.1: e0220019. pmid:31945053
- 34. PHILLIPS Caleb M., et al. A hybrid model of tumor growth and angiogenesis: In silico experiments. Plos one, 2020, 15.4: e0231137. pmid:32275674
- 35. WINITCHAIKUL Thanwarat, et al. Calotropis gigantea stem bark extract induced apoptosis related to ROS and ATP production in colon cancer cells. PLoS One, 2021, 16.8: e0254392. pmid:34343190
- 36. WONG Kah Keng; HUSSAIN Faezahtul Arbaeyah. TRPM4 is overexpressed in breast cancer associated with estrogen response and epithelial-mesenchymal transition gene sets. PLoS One, 2020, 15.6: e0233884. pmid:32484822