In recent years, extensive research has been done to reduce fault detection and recognition of breast cancer masses and increase speed and accuracy to assist radiologists. In general, research in this area; Includes sections for selecting the appropriate database, including digital mammography images (images of healthy tissue, and tissue with benign and malignant masses), image preprocessing, identification and extraction of parts of breast tissue containing cancerous masses, expansion of the susceptible area and also the mass has been exposed to the boundaries of the mass or vice versa. Hence, first, the boundaries of the mass are revealed and then the final diagnosis of the image containing the cancerous mass is carried out. Various features (such as morphological, geometric, tissue-based, violet, etc.) are extracted from the mass, and subsequently, the final classification of the extracted properties is performed to determine the type of cancerous mass (benign or malignant) by intelligent algorithms. In the continuation of this section, some of the most important researches in this field are introduced.
Breast cancer is one of the main causes of death for women. Recent studies have revealed that the best way to prevent cancer is through routine screening and prompt treatment. X-ray mammography can be considered as an early diagnostic procedure. Good and sometimes very gentle contrast of healthy and unhealthy lymph node tissues has been described to help treat radiologists and their internal medicine specialists. A new approach is presented in diagnosing mammograms with weight in breast cancer (Ghantasala et al., 2020). The proposed study of the morphological operators is noticeably identified for the segmentation and clustering of existing anomalies in the form of biomass and microcalcification. The findings indicate that most of the abnormalities are recognized and eliminated for the lower gray comparison points; however, there are other areas with the same texture. Certain areas with identical textures, on the other hand, gradually vanish from the image for higher gray reference values, although the area of distinct disturbance and discrimination is against a smaller area. Therefore, the appropriate value of the gray reference level is required to ensure that the suspected regions are effectively divided and extracted, and efficiently classified, thereby preventing the extraction of areas unrelated to the same tissues is minimized. The optimal value of the gray reference point in this regard is shown to be calculated about to with concerning the overall image size as a result of the relative error evolution. However, this cannot be a safer option in terms of medical recognition assistance because sensitive areas associated with non-relevant areas can be identified instead of being ignored and then deprived of a significant sequence of the potential pathological sites. The planned algorithm enables reliable and scalable computing only by changing the grey reference point to a given threshold value.
A method of enhancing mammograms for visual interpretation of the breast masses was suggested using the measurement of adaptive neuro-fuzzy divergence under Hyperbolic Regularization (HR) (Ghosh and Ghosh, 2022). The proposed scheme is developed to increase the unpleasant and abnormal growth of cells such as breast masses, abnormal tissues, nodules, and masses in mammography images. In the first stage, a complementary image of mammography is produced to separate the object and the background. Then, both source and background images are imaged with Intuitionistic Fuzzy Set (IFS) under Hyperbolic Regularization (HR). A new entropy-based fuzzy deviation criterion is designed using the hyperbolic function to modify the membership. Moreover, the distance function of the fuzzy ambiguity interval is obtained from the vectors of hesitation from both the source and the complement images. Werner’s AND-OR has been used in both to generate a modified membership function. Finally, by improving the quality of the mammogram, an increase in breast masses is achieved through the defuzzification process.
A computer method has been presented recently for the segmentation of microcalcifications in mammograms (Ciecholewski, 2017). This makes use of morphological changes and consists of two parts. The first section recognizes the microcalcification from the morphological point of view, allowing the approximate range of occurrences to be determined, the contrast improves and, the noise in mammograms decreases. In the second part, watershed classification of microcalcifications is accomplished. This study was performed on an experimental set containing 200 pixels 512 × 512 × 512 × 512 Rois taken from mammography from the Digital Database for Mammography Screening (DDSM), including 100 malignant lesions and 100 benign lesions. The performed tests were averaged to obtain the following values of the measured indices: 80.5% (similarity index), 75.7% (overlap fraction), 80.8% (overlap value) and 19.8% (additional fraction). The average execution time of all steps of the methods used for ROI is about 0.83 seconds.
An automated binary model for tissue diagnosis in digital mammograms is suggested as a support tool for radiologists (Fanizzi et al., 2020). For each ROI, texture features are identified on HAAR wavelet decomposition as well as points of interest and corners identified using the Speeded Up Robust Feature (SURF) and EigenValue (Mineigenalg). Binary classification of random forest is then trained on a subset of features selected by two different types of feature selection techniques such as filtering and embedded methods. It should be noted that the proposed model was tested on 260 ROIs extracted from digital mammograms of the BCDR public database. The best predictive performance for normal/abnormal and benign/malignant problems is the average AUC of 98.16% and 92.08% and the accuracy of 97.31% and 88.46%, respectively.
Since the radiologist extensively uses mammography as the primary means of screening for breast cancer, the exact diagnosis of Microcalcifications is an unavoidable step to develop an effective diagnosis system. A Stationary Wavelet Transform (SWT) technique has been presented for the detection and classification of breast microcalcifications (Mazumder et al., 2020). To identify suspicious regions of mammography, SWT was performed at different levels for the purpose of constant wavelet energy analysis to extract specificity from each of the obtained SWT coefficients. Four different group classifications were utilized to categorize Microcalcifications as benign or malignant using these SWE features, performing 10 cross-validations.
The survey (Kang et al., 2021) was carried out with the objective of assessing diagnostic performance using deep convolutional neural network (DCNNS) in the Microcalcifications classification of the breast at the mammograms. For this purpose, 1579 mammographic images were collected from patients with suspected microcalcifications in screening mammography between July 2007 and December 2019. Five DCNN models were previously used to classify Microcalcification as malignant or benign. Nearly one million images from the ImageNet database have been used to teach the five DCNN models. Hence, 1121 mammographic images were used for individual model adjustment, 198 for validation and, 260 for testing. Gradient-weighted classification activation mapping (Grad-Cam) was used to authorize the validity of DCNN models in highlighting Microcalcification areas to determine the final class.
Two new methods (Christopher and Simon, 2020) are presented for improving mammogram images based on a new Unsharp mask design, called nonlinear minimum and slope minimization, NLUM GMIn (nonlinear UM and L0 Gradient Minimization, NLUM GMIN) for cancer diagnosis. Three different techniques are combined in the proposed method; I) a non-linear filtering process to increase precise detail in a local 3×3 neighborhood; II) A globally minimized L0 gradient step that preserves high-contrast edges while suppressing low-contrast detail as matte fiber textures; and obtain a partial mammogram after subtracting a smooth mammogram from the original mammogram, III) Finally, the Unsharp mask technique combines the details of filtered mammography through a non-linear filter, using PLIP operators that meet both Weber’s law features and the saturation features of the human visual system. An HVS-based analysis scheme is used to analyze and visualize malignant areas on advanced mammography. The distinct arrangement of Plip operators in the proposed framework leads to the NLUM Gminauto and NLUM0Gmin methods, which use the NLUM method and other available techniques to increase the mammogram. The results indicated that NLUML0Gmin's proposed scheme for detecting cancerous masses and microcalcifications in dense X-ray mammograms is strong and effective in assisting physicians in better diagnosing cancer by improving the lives of countless cancer patients.
The correlation between two-dimensional shear-wave elastography (2D-SWE) and histopathological results of microcalcifications (MCS) was performed using ultrasonography (USG) (Kayadibi et al., 2021). Fifty people with suspected MCS were evaluated. They were monitored with USG and 2D-SWE before Tru-Cut biopsy. It is worthy to note that SWE values and histopathological features were compared statistically. Variables between groups were analyzed using the Mann Whitney U test. Consequently, SWE is a useful approach in clinical practice to identify MCS that can be visualized with USG.
Automatic detection of microcalcification is essential for proper diagnosis and treatment. An automated approach and parts of microcalcification are introduced in mammographic images (Hossain, 2019). Initially, to enhance the image, image preprocessing programs are applied. Subsequently, the breast area is separated from the Pectoral area. Suspicious areas are distinguished using the C-means fuzzy clustering algorithm and divided into negative and positive sections. This method eliminates the need to manually tag the area of interest. Positive snippets containing microcalcification pixels are taken to teach a modified U-Net segmentation network. Finally, the trained network is used to automatically isolate the microcalcification area from mammographic images. This procedure can be served as an assistant radiologist for early detection and increase the accuracy of segmentation of microcalcification areas.
A method has been presented for mammographic imaging using Fuzzy rough set theory (FRST) and a detailed approach to texture and feature extraction (Punitha and Perumal, 2019). The main objective of the FRST setup is to extract the feature that is obtained using a rapid reconstruction algorithm which is extracted. The main purpose of establishing FRST is to extract the feature, which is obtained using a rapid regeneration algorithm that helps identify the tumor without losing pixels in a short period time. Fuzzy rough instance selection (FRIS) is utilized to eliminate noise from the mammographic image, and finally, a combination of the fuzzy-rough nearest neighbor (FRNN) method is used in the segmentation.
A three-step method for the detection of cluster microcalcification is planned (Basile et al., 2019). In particular, it is made up of a preprocessing step, aimed at highlighting potentially interesting breast structures, followed by a single microcollection detection step, based on the Hugh Hough transform, which can be able to understand the specific shape of interest structures. Finally, a cluster identification step is performed for the Microcalcifications group using clustering algorithm capable of implementing expert domain rules. An automated approach has been performed to detect the location of each microcalcification in mammographic images in a complete and simple manner (Hakim et al., 2021). First, the image processing algorithms were applied to increase the image quality. Subsequently, the microcalcification area was labeled using the image segmentation based on the radiologist’s expertise. Positive tags containing microcalcification pixels are taken for training with a segmentation grid. In this research study, 354 images from InBreast data were used. Finally, to automatically detect the area of microcalcification from mammographic images, the trained network was utilized.
Calcification was characterized by descriptors derived from deep learning and bracelet descriptors. they compared the performance of different image sets to digital mammography (Cai et al., 2019). (E sets of attributes include profound features alone, manipulated features, combination of them, and deep properties). Experimental results have revealed that deep features outperform manipulation features, but manipulation features can provide complementary information for deep features. In this effort, the classification accuracy is 89.32% and the sensitivity is 86.89% using deeply filtered features, which can be considered as the best performance among all feature sets.
Two approaches to feature extraction using Empirical Mode Decomposition (EMD) are proposed to classify masses in mammographic images as benign or malignant (Nagarajan et al., 2019). The first method of feature extraction is based on Bi-dimensional Empirical Mode Decomposition (BEMD). In performing BEMD in the Region of Interest (ROI) mammogram, the ROI is broken down into a set of different frequency components called Bi-dimensional Intrinsic Mode Functions (BIMFS). Gray Level Co-occurrence Matrix (GLCM) and Gray Level Run Length Matrix (GLRM) are extracted from these BIMFs and given as input for benign or malignant classification. Due to the mixing mode problem that exists in BEMD, the BIMFS obtained from BEMD are less orthogonal to each other. To overcome this, the second method of feature extraction is proposed in the name of Modified Generator Experimental State Decomposition (MBEMD). BIMFs are extracted using the proposed MBEMD in ROI mammography. Properties are extracted in the same way as the BEMD method. Support Vector Device (SVM) and Linear Discriminant Analysis (LDA) are used to classify mammogram mass.