Automatic detection of breast cancer in ultrasound images using Mayfly algorithm optimized handcrafted features

Abstract

BACKGROUND:

The incidence rates of breast cancer in women community is progressively raising and the premature diagnosis is necessary to detect and cure the disease.

OBJECTIVE:

To develop a novel automated disuse detection framework to examine the Breast-Ultrasound-Images (BUI).

METHODS:

This scheme includes the following stages; (i) Image acquisition and resizing, (ii) Gaussian filter-based pre-processing, (iii) Handcrafted features extraction, (iv) Optimal feature selection with Mayfly Algorithm (MA), (v) Binary classification and validation. The dataset includes BUI extracted from 133 normal, 445 benign and 210 malignant cases. Each BUI is resized to 256×256×1 pixels and the resized BUIs are used to develop and test the new scheme. Handcrafted feature-based cancer detection is employed and the parameters, such as Entropies, Local-Binary-Pattern (LBP) and Hu moments are considered. To avoid the over-fitting problem, a feature reduction procedure is also implemented with MA and the reduced feature sub-set is used to train and validate the classifiers developed in this research.

RESULTS:

The experiments were performed to classify BUIs between (i) normal and benign, (ii) normal and malignant, and (iii) benign and malignant cases. The results show that classification accuracy of > 94%, precision of > 92%, sensitivity of > 92% and specificity of > 90% are achieved applying the developed new schemes or framework.

CONCLUSION:

In this work, a machine-learning scheme is employed to detect/classify the disease using BUI and achieves promising results. In future, we will test the feasibility of implementing deep-learning method to this framework to further improve detection accuracy.

Keywords

Breast cancer ultrasound image handcrafted features mayfly algorithm binary classification

1 Introduction

In the current era, the incidence rate of various infectious and acute diseases in humankind and appropriate diagnosis and treatment is the only option to reduce the impact of the disease [13 , 35]. The medical imaging supported disease diagnosis is a suggested practice in hospitals and hence a number of medical imaging procedures are developed and implemented to detect the disease with better accuracy [18 , 29]. Ultrasound Imaging Technique (UIT) is one of the approaches in which the necessary images are captured using sound wave. Compared to other imaging schemes, the UIT is considered to be safe and it can be considered to capture various information as discussed in [6].

Cancer is recognized as one of the harsh diseases in humans and premature detection is necessary to control its growth and spread [3 , 33]. The statement by World Health Organization (WHO) confirms that cancer has a wider group and it can happen in any organ/tissue of body. This report also confirms that 9.6 million deaths happened globally in 2018 [34]. This is mainly due to the abnormal cell growth and the grown cells will spread to other parts of the body through the blood stream. Breast Cancer (BC) is one of the leading causes of death in women and the WHO report confirms that, in early 2020 about 2.3 million women were diagnosed with BC and 685000 reported deaths globally. Further this report also confirms that, at the end of year 2020, 7.8 million women living globally have BC confirmed within five years.

Cancer signs/symptoms rely upon various factors like cancer type, its location or the neighboring area where it has spread. For example, breast cancer can appear as a lump within the breast or some sort of discharge from nipple while breast cancer that is metastatic may project pain signals (if bones are also affected), excessive fatigue (linked with lungs), or seizures (in brain).

Because of its happening rate and harshness, a considerable number of diagnostic and treatment procedures are proposed and employed to detect the BC in its premature stage. The BC caused a large medical burden globally and to support early detection and treatment, a number of diagnosis procedures are developed and employed. Medical imaging supported BC detection is commonly adopted in hospitals and schemes, such as Magnetic-Resonance-Imaging (MRI), mammogram, thermal imaging, histopathology, elastography and UIT are widely employed in hospitals. Compared to other methods, the UIT is considered to be safe and helps to detect the location of the cancer during the needle biopsy procedure. The UTI or UTI combined with elastography is widely adopted in hospitals to diagnose the BC with better accuracy.

The most reported challenge by most of the breast cancer classification and segmentation mechanism is regarding the accurate segmentation and classification of ultrasound region, which is addressed in this research through exploiting the effective breast cancer segmentation method. Today, breast cancer portrays a frightening picture in every individual’s mind as it is considered to be intolerable, agonizing and fatal. In fact, this outlook has been overstated and amplified baselessly.

Breast Cancer (BC) generally begins with cells in the milk-producing ducts. It may also start in the glandular tissue called lobules or in other cells or tissue within the breast. An early detection and treatment will help to cure the disease.

In the literature, a number of BC diagnostic procedures are implemented with medical images of various modalities and in this work, the Breast Ultrasound Image (BUI) is considered for the investigation. Table 1 summarizes the earlier works employed on BUI supported cancer detection.

Table 1
Summary of ultrasound supported breast cancer detection

Reference Methodology

Dey et al. [2] Computerised detection and classification of BC with BUI is presented with appropriate technique.

Giuliano and Giuliano [4] Detection of the breast abnormality in dense breast is implemented and discussed with 3D BUI.

Kadry et al. [7] A detailed discussion about the BC detection and handling is presented.

Kelly et al. [10] Detection of the breast abnormality using mammography along with/without the ultrasound is discussed and this work confirmed that the ultrasound assisted approach provides better diagnosis.

Kozegar et al. [12] This work presented the detection of the breast cancer in 3D BUI using the iso-contours combined with cascaded RUSBoosts.

Ozmen et al. [20] This work presented an analysis on various BUI, such as time-of-flight tomography, synthetic aperture focusing, back-propagation, Born inversion, and contrast source inversion methods employed in BC diagnosis.

Shi et al. [26] Detection of masses in BUI is implemented with a machine learning technique and fuzzy SVM classifier.

Sood et al. [28] A detailed review of various techniques employed to detect the BC with BUI is presented.

Reference	Methodology
Dey et al. [2]	Computerised detection and classification of BC with BUI is presented with appropriate technique.
Giuliano and Giuliano [4]	Detection of the breast abnormality in dense breast is implemented and discussed with 3D BUI.
Kadry et al. [7]	A detailed discussion about the BC detection and handling is presented.
Kelly et al. [10]	Detection of the breast abnormality using mammography along with/without the ultrasound is discussed and this work confirmed that the ultrasound assisted approach provides better diagnosis.
Kozegar et al. [12]	This work presented the detection of the breast cancer in 3D BUI using the iso-contours combined with cascaded RUSBoosts.
Ozmen et al. [20]	This work presented an analysis on various BUI, such as time-of-flight tomography, synthetic aperture focusing, back-propagation, Born inversion, and contrast source inversion methods employed in BC diagnosis.
Shi et al. [26]	Detection of masses in BUI is implemented with a machine learning technique and fuzzy SVM classifier.
Sood et al. [28]	A detailed review of various techniques employed to detect the BC with BUI is presented.

Table 1 presents the necessary information existing in the literature on the BC detection with the BUI and these works confirms its merit over other modalities. Compared to the non-invasive imaging approach, like thermal imaging, the recording and evaluation of the BC in BUI is quite simple and hence, it is commonly adopted in hospitals to detect the abnormalities. In this work, computerized classification of the BUI is presented using the benchmark images collected from [1]. The main limitations in the earlier work are the choice of the features which offered a lesser detection accuracy with the Machine-Learning Schemes (MLS). The proposed experimental investigation is implemented using the benchmark images contributed by Al-Dhabyan et al. [1]. In this database, every image is associated with the Ground-Truth and the collection, pre-processing and post processing of the images are clearly discussed. In this study, the a image resizing is implemented initially to reduce the test images into 256×256×1 pixels and the image augmentation (rotation with angle) is implemented to increase the number of test images (500 numbers) as depicted in Table 2. During the classification task, 70% of these images are considered for training the MLS and 30% of images are considered for the validation. The sample pictures of this database are presented in Fig. 1. During this investigation, a separate examination is employed for (i) Normal Vs Benign, (ii) Normal Vs Malignant and (iii) Benign Vs Malignant and the results are compared and verified using various binary classifiers available in the literature. The experimental outcome of the proposed scheme confirms that the developed MLS is efficient in classifying the considered BUI with better accuracy.

Table 2

Sample test images considered in this research

Class	Dimension	Images
		Actual	Augmented	Training	Validation
Normal	256×256×1	133	500	350	150
Benign	256×256×1	445	500	350	150
Malignant	256×256×1	210	500	350	150

Fig. 1

Sample test pictures from chosen BUI database.

The other portions of this manuscript are arranged as: Portion 2 presents the context, Portion 3 illustrate the methodology, Portion 4 and 5 demonstrate results and conclusions, respectively.

2 Methodology

This research work proposed a Machine Learning System (MLS) to detect the Breast Cancer (BC) and the proposed scheme is depicted in Fig. 2.

Fig. 2

Developed scheme to examine the breast abnormality.

This technique consists the following phases; collecting the images of normal/abnormal breast section from the volunteers using the ultrasound imaging scheme, resizing the collected images to a recommended dimension (256×256×1), enhancing the images with a chosen procedure, extracting the vital features from the enhanced images, feature optimization and classification with dominant features. The most common classification methods are voxel-based or pixel-based breast cancer classification techniques. However, these approaches failed to consider the boundary information and global shape of breast cancer detection [5].

The performance of this scheme relies majorly on the feature mining and selection procedure. In this work, the test images are enhanced using the methods, such as Gaussian filter, LBP and saliency detection technique and the necessary features are then extracted from these images. The extracted features are then optimized with Mayfly-Algorithm (MA) and these features are then considered to validate the performance of the proposed scheme.

The gaps and issues identified from the optimization based methods are as follows: The detection rate was very less in [9], the performance was degraded with the increasing training samples and the computation resource required was also too high. In Vijayakumar et al. [21], the efficiency of the segmentation method was poor. It failed to attain better performance for the entire breast cancer region in Shi et al. [26] and the complexity and multiplicity of breast cancer made the process more tedious [8]. The memory load and computation overhead were increased in Dey et al. [2] but it failed to model the 3D network for enhancing the segmentation performance. The detection performance for breast cancer was less in Rajakumar et al. [21]. In case of white matter region, this model recognizes only the intensities of texture space, and faced a complex issue in computing the weight distance based on the breast cancer voxels [3]. Detecting and preventing breast cancer at the earliest, can be beneficial in minimizing the death rate and increasing the life span. Additionally, with early detection, the prognosis can be improvised, resulting in the best appropriate and effective cure.

Usually, Breast cancer formation can initiate from any region of the breast. The ductal cancers initiate from the ducts which carries milk to the nipple. The lobular cancers initiate in the glands that produce breast milk. Much other unfamiliar breast cancer also prevails [23]. There is also some tiny form of cancer that begins in some region of the breast tissues. These are referred to as sarcomas and lymphomas which are actually not considered as breast cancers. Though a lump in the breast can be formed because of any sort of cancer, it is not actually formed by all [1].

Various Breast cancers are identified on screening mammograms, which identify cancers at quiet an early stage, even prior to their feeling or any symptoms. Breast cancer has various other warning signs which one must not overlook and need to be conveyed to the doctor. One must also comprehend the fact that most of the breast lumps are benign rather than malignant. Breast tumors that are non-cancerous are a merely malformed growth that does not permeate outside the breast region and are non-hazardous [11].

Though there exist some benign breast lumps that can increase the probability of breast cancer in women. On experiencing any lump in the breast or any unusual change, the same must be immediately reported to the doctor to verify the existence of any benign or malignant cells and any risk involved in contracting cancer in the near future [24].

It is essential that the clinicians and oncologists identify particular risk parameters related to cancer disease for assisting in adjuvant therapy and thereafter surgical therapy considering overall elimination of affected region as and when required. Cancer diagnosis and pathological analysis of breast cancer affected regions are illustrated in experiments for providing effective and valuable survivorship advantage [22]. Hence it is necessary that patients undergoing cancer diagnoses must be given precise and overall diagnostic evaluation which will act as a foundation for the novel therapeutic strategies. Multiple scoring based diagnosing systems are examined. For complementing the diagnosis conducted by the medical experts, various researches are carried out for the growth and advancement of diagnosis systems which are presented in the subsequent section.

2.1 Image enhancement

The performance of the automated disease detection scheme relies mainly on the features obtained from the images to be categorized. In Machine-Learning technique, a number of feature extraction methods are discussed in prior works and in this research, the necessary features are extracted from the images enhanced with GF, LBP and saliency. The necessary entropy features, such as Kapur, Tsallis, Shannon, Yager, Rényi and Max are extracted from the images of GF and Saliency. Along with these entropies, three Hu moments are extracted from every image. The LBP helps to get 59 features from each pattern and these images are then optimized using the MA. This section presents the overview of the image enhancement techniques implemented in this work.

2.1.1 Gaussian filter

Surface and edge improvement provide essential image information and the GF based picture enhancement is a proven technique. The work of Marr and Hildreth [15] authenticates that the GF implemented using varied scale (θ) improves the texture pattern in vertical/horizontal orientations. The expression for 2D Gaussian operator is defined in Eqn. (1); $G (X, Y) = \frac{1}{2 π θ^{2}} e^{- (\frac{X^{2} + Y^{2}}{2 θ^{2}})}$ (1) where θ is the standard deviation of the Gaussian value and G (X, Y) is image Cartesian coordinates. In this work the GF pattern developed with various θ is shown in Fig. 3. These patterns are considered to extract the entropy and Hu values.

Fig. 3

GF patterns generated with various θ.

The GF helps to get 6 entropy values and three Hu moments from each image. The total features extracted from four patterns are presented in Equation. (2) and final value is shown in Equation (3); $G F_{(1 \times 1 \times 9)} = Entrop y_{(1 \times 1 \times 6)} + H u_{(1 \times 1 \times 3)}$ (2) $G F_{(1 \times 1 \times 36)} = 4 \times (Entrop y_{(1 \times 1 \times 6)} + H u_{(1 \times 1 \times 3)})$ (3)

2.1.2 Local Binary Pattern

The LBP based information extraction is recently considered in medical image classification tasks and the earlier work confirms that, this technique helps to get 1 × 1 ×59 features from each image.

In this work, the LBP with various chosen weights proposed in Gudigar et al. [5] is considered and the generated patterns for adopted weights of W = 1 to 4 is shown in Fig. 4. Other information can be found in [2].

Fig. 4

LBP patterns generated using different weights.

The number of mined LBP features is presented in Equation (4); $LB P_{(1 \times 1 \times 236)} = LBP 1 + LBP 2 + LBP 3 + LBP 4$ (4)

2.1.3 Saliency

Saliency (SAL) is associated with the enhanced section from an image and the saliency detection will help to identify the vital information present in the test image. The earlier work on salience is discussed in [11]. In this work, the saliency in BTU is computed to identify its abnormal section and the hot colour map outcome during this task is depicted in Fig. 5. This image helps to get the feature presented in Equation (5); $SA L_{(1 \times 1 \times 9)} = Entrop y_{(1 \times 1 \times 6)} + H u_{(1 \times 1 \times 3)}$ (5)

Fig. 5

Saliency value of a sample test image.

After collecting the necessary features from the pictures, every feature is integrated to form the necessary feature vector and this feature vector is then considered for the feature reduction with MA. The overall features collected from the image enhancement task are depicted in Equation (6); $Featur s_{(1 \times 1 \times 281)} = G F_{(1 \times 1 \times 36)} + LB P_{(1 \times 1 \times 236)} + SA L_{(1 \times 1 \times 9)}$ (6)

2.2 Feature optimization

After performing the segmentation process, the next step to be performed is the feature extraction. Relevant and the essential features are extracted from the segmented ultrasound image. Here, the features from the ultrasound image are extracted using information theoretic measures. To extract the features, the original breast ultrasound image and the generated segments are employed such that the original image features along with the features assures better classification accuracy. The theoretic measures enhance the accuracy of classification by offering texture features from the classifier.

Feature optimization is the main task in every MLS to avoid the problem of over fitting in [33]. The traditional (Student’s t-test) assisted feature reduction procedure needs various mathematical operations to be employed and this work also governed by p-value. To minimize the complexity in the feature reduction/optimization task; heuristic algorithm-based techniques are widely employed in the literature and this work adopted the MA based feature reduction discussed by Rajakumar et al. [21].

The MA is a recently invented heuristic technique [7, 36], which helps to find the optimal solution based on the Cartesian-Distance (CD) existing among the search agents. The operation of MA is similar to the traditional Firefly Algorithm and it provides the optimal solution with lesser iterations. The chief merit of the MA is it consist the merits of Particle Swarm Optimization (PSO), Firefly Algorithm (FA) and Genetic Algorithm (GA) and hence it provides a better result.

During examination stage, all agents are permitted to connect close to the premium location (G_best). Later, M is permitted to join at (G_best) by altering its location and speed. This process is shown in Equations (7), (8) and (9); $P_{i}^{t + 1} = P_{i}^{t} + V_{i}^{t + 1}$ (7) $V_{i, j}^{t + 1} = V_{i, j}^{t} + C_{1} * e^{- β D_{p}^{2}} (p_{bes t_{i, j}} - P_{i, j}^{t}) + C_{2} * e^{- β D_{g}^{2}} (G_{bes t_{i, j}} - P_{i, j}^{t})$ (8) where $P_{i}^{t}$ and $P_{i}^{t + 1}$ are initial and final locations, $V_{i}^{t + 1}$ and $V_{i, j}^{t + 1}$ initial and final velocities, respectively. C₁ = 1 and C₂ = 1.8 signifies local and global learning values. β = 2, D_p and D_g are the CD.

The velocity update by; $V_{i, j}^{t + 1} = V_{i, j}^{t} + d * R$ (9) where nuptial-dance (d) = 5 and R = arbitrary digit [–1,1].

The velocity and position update for the F is shown in Equations (9) and (10); $P_{i}^{' t + 1} = P_{i}^{' t} + V_{i}^{' t + 1}$ (10) $F_{i, j}^{' t + 1} = {\begin{matrix} F_{i, j}^{' t} + C_{2} e^{- β D_{mf}^{2}} (M_{i, j}^{t} - Y_{i, j}^{t}) if O (F_{i}) > O (M_{i}) \\ F_{i, j}^{' t} + W * r if O (F_{i}) \leq O (M_{i}) \end{matrix}$ (11) where O= maximized CD.

Figure 6 depicts the stages available in the MA and it’s working. Here, the optimization capability and the ability of feature extraction were joined together to perform effective breast cancer detection. The large convolutional kernels and the small convolutional kernels structure were combined to reduce the over fitting and to increase the nonlinear mapping such that the features with the multiformity rate was also increased. The features that were learned from the ultrasound image were fed to the classifier for implementing the optimization framework.

Fig. 6

Flow chart of the MA based feature selection.

The feature selection operation is graphically presented in Fig. 7; and this scheme helps to reduce the feature vector to a lower level. In this scheme, every feature (Normal/Cancer) is individually compared with its related feature and the feature with better CD is selected and the worst CD is discarded. The selected features are then combined together to form a 1D feature vector, known as optimal features.

Fig. 7

Feature reduction with MA by maximizing the CD.

2.3 Classification and Validation

By employing Machine learning a program can examine statistics, comprehend correlations and utilize the insight for resolving issues and improvise predictions/facts. Techniques of data mining and machine learning algorithms hold a very important stand in the medical domain. Health informatics has evolved and is growing at a fast pace dealing with computer /Information technology and Information technology in the field of health and clinical information. The approach of Clinical data mining suffers from some sort of misdiagnosis and uncertainty.

The initial parameters of these classifiers are tuned as per the discussion in the earlier works and in a few algorithms, the traditional tuning parameters existing are considered unchanged. The necessary information about these classifiers can be found in [9, 11].

The performance of developed BUI assessment scheme is confirmed by computing the necessary image quality parameters. Based on these values, other measures depicted in Equations (12) to (17) are calculated and these values confirm the performance of proposed scheme. $Accuracy = \frac{TP + TN}{TP + TN + FP + FN}$ (12) $Precision = \frac{TP}{TP + FP}$ (13) $Sensitivity = \frac{TP}{TP + FN}$ (14) $Specificity = \frac{TN}{TN + FP}$ (15) $F 1 Score = \frac{2 T P}{2 TP + FN + FP}$ (16) $Negative Predictive Value = \frac{TN}{TN + FN}$ (17) where TP, TN, FP and FN represent true-positive, true-negative, false-positive and false-negative, respectively.

3 Results

This section of the research presents the experimental outcome of the present study. This work is tested using a workstation equipped with MATLAB software and the images present in Table 2 are considered for the assessment.

Initially, every image is treated using the GF with a chosen angle (θ = 30^o, 120^o, 240^oand 300^o) and the obtained images are recorded. After generating the GF patterns, the features are then extracted using chosen Entropies and Hu moments and are stored as the GF features. In the next phase, the test images are then improved using the LBP with weights, like W = 1 to 4 and the necessary LBP features (1 × 1 ×236) are then extracted and considered. Finally, the saliency of the images is generated and the Entropies as well as Hu moments are extracted.

Figure 8 presents the outcome achieved for various test images (Normal, Benign and Malignant class BUI). This image confirms that the texture pattern of the image various based on the enhancement methods and the features extracted from these images will provide the necessary insight about the information to be analysed. All these extracted features are serially combined to get a single vector of size 1 × 1 ×281; then a feature reduction is implemented to extract the vital features using the MA, which provides a reduced 1D feature vector of size 1 × 1 ×94.

Fig. 8

Enhanced test image with various methods.

The feature optimized by the MA is then considered to train and validate the binary classifiers with a 5-fold cross validation. This comparison helps to get the necessary results and based on these values, the performance values are compared and validated. In this work binary classification is implemented to categorize the BUI into Normal Vs Benign, Normal Vs Malignant and Benign Vs malignant and the achieved results are presented in Table 3. This table initially computes the TP, FN, TN and FP values and from these values, other measures are derived in [8].

Table 3

Performance values achieved during experimental analysis

Approach	Classifier	T_+ve	F_ve	T_ve	F_+ve	Accuracy	Precision	Sensitivity	Specificity	F1Score	NPV
Normal Vs Benign	NB	138	12	136	14	91.3333	90.7895	92	90.6667	91.3907	91.8919
	RF	137	13	139	11	92	92.5676	91.3333	92.6667	91.9463	91.4474
	KNN	142	8	142	8	94.6667	94.6667	94.6667	94.6667	94.6667	94.6667
	SVM-L	142	8	144	6	95.3333	95.9459	94.6667	96	95.3020	94.7368
	SVM-RBF	143	7	141	9	94.6667	94.0789	95.3333	94	94.7020	95.2703
Normal Vs Malignant	NB	140	10	142	8	94	94.5946	93.3333	94.6667	93.9597	93.4211
	RF	139	11	137	13	92	91.4474	92.6667	91.3333	92.0530	92.5676
	KNN	145	5	142	8	95.6667	94.7712	96.6667	94.6667	95.7096	96.5986
	SVM-L	142	8	143	7	95	95.3020	94.6667	95.3333	94.9833	94.7020
	SVM-RBF	142	8	141	9	94.3333	94.0397	94.6667	94	94.3522	94.6309
Benign Vs Malignant	NB	139	11	135	15	91.3333	90.2597	92.6667	90	91.4474	92.4658
	RF	140	10	143	7	94.3333	95.2381	93.3333	95.3333	94.2761	93.4641
	KNN	142	8	140	10	94	93.4211	94.6667	93.3333	94.0397	94.5946
	SVM-L	139	11	137	13	92	91.4474	92.6667	91.3333	92.0530	92.5676
	SVM-RBF	141	9	140	10	93.6667	93.3775	94	93.3333	93.6877	93.9597

In this research, the features of 350 images are considered to train the classifier system and the features of 150 images are used for the validation. For Normal Vs Benign case; the classification accuracy achieved with SVM-L is better compared to other methods. Individual comparison of the classifier performance is time consuming and hence, the overall performance of this technique is compared with the help of Glyph-Plot as depicted in Fig. 9(a). The pattern which covers the largest area is considered to be better and from this figure, it is confirmed that the overall performance of SVM-L is good (> 95%) compared to other binary classifiers considered in this research work.

Fig. 9

Glyph-Plot constructed using the outcome of the proposed scheme.

Similar process is then executed for Normal Vs malignant and Benign Vs Malignant and the related results are depicted in Table 3 and Fig. 9(b) and (c) respectively. The outcome of the KNN is better (> 96%) compared to the alternatives in the first case and the result by RF is better (> 94%) in the second case. These results confirm the merit of the proposed scheme and this confirms that, the present approach can be considered to examine the breast abnormality using the BUI.

4 Discussions

Figure 9 presents the comparison of the accuracy achieved with the proposed scheme for various classes of images and this confirms that the accuracy achieved with Normal Vs Malignant class is better compared to other comparisons. In most of the test images, the features extracted with the normal and benign class pictures are alike and hence the classification accuracy if less compared to the malignant class. Further, when comparing the malignant class with the benign class, getting better accuracy is reduced due to the alike GF, LBP and Saliency features; which look similar in most of the images. From Fig. 10, it can be confirmed that, the proposed scheme helps to get enhanced disease detection when we consider normal and malignant class images (with KNN classifier) and slightly lesser detection when we consider the benign class image with normal/malignant class images.

Fig. 10

Comparison of achieved accuracy with proposed scheme with various classifiers.

In order to validate the merit of the proposed scheme, its classification accuracy is then compared with other BTI examination procedures existing in the literature as shown in Fig. 11. In the literature, similar data set is considered to detect the cancer using the methods, like Convolutional Neural Network (CNN), SVM [19] and Artificial Neural Network (Singh et al., 2015) and the accuracy is then compared with the proposed technique,

Fig. 11

Comparison of accuracy in proposed and existing methods.

The implemented ensemble learning with CNN and achieved 94.62% accuracy. The work of Nugroho et al. [19] executed texture and geometry-based disease detection and the SVM classifier provided accuracy of 91.30%. The research by Singh et al. [15] considered texture and shape information and with ANN classifier, an accuracy of 84.60% is obtained. When comparing to other earlier works, the proposed scheme helps to get a better accuracy as shown in Fig. 11.

The outcome in this research confirms that the proposed scheme works well on the considered BTI database. This scheme considered the well-known features to get the better detection accuracy. In future, this work can be extended with; (i) Considering other texture and shape features to enhance the outcome, and (ii) Combining the machine-features with deep-features to improve the disease detection accuracy. In future, the performance of the proposed scheme can be tested and confirmed using the clinically collected BTI from hospitals.

5 Conclusion

This proposed research applies a machine-learning technique to detect the cancer in BUI with better accuracy. This work considered the test images from the benchmark BUI database and every test image of this dataset is resized to 256 × 256 × 1 and this image is then enhanced with texture enhancement techniques. Every image is treated using GF, LBP and Saliency techniques and from each image, the necessary features, such as entropies, Hu and LBP are then extracted. To avoid the over fitting problem, a MA based feature reduction is then employed, which reduced the actual feature 1 × 1 ×281 to 1 × 1 ×94. This scheme employed binary classification to detect the disease in Normal Vs Benign, Normal Vs Malignant and Benign Vs Malignant classes and this technique helped to get a classification accuracy of > 94%. This research focuses on the Mayfly optimization based on the machine learning technology, this can be further extended to use some other optimization to achieve effective performance is severity level breast cancer classification. Also, this work can be extended to minimize the computational overhead.

Conflict of interest

The authors declare that they have no conflict of interest.

Footnotes

Acknowledgments

The authors of this research work thank Dr. Aly Fahmy for sharing the database for the research work.

Data Availability

The image database considered in this work can be accessed from;

References

Al-Dhabyani

, Gomaa

, Khaled

, Fahmy

Dataset of breast ultrasound images, Data in Brief 28 (2020), 104863.

Dey

, Rajinikanth

, Fong

S.J.

. et al., Social group optimization–assisted Kapur’s entropy and morphological segmentation for automated detection of COVID-19 infection from computed tomography images, Cognitive Computation 12(5) (2020), 1011–1023.

Fernandes

S.L.

, Rajinikanth

, Kadry

A hybrid framework to evaluate breast abnormality using infrared thermal images, IEEE Consumer Electronics Magazine 8(5) (2019), 31–36.

Giuliano

, Giuliano

Improved breast cancer detection in asymptomatic women using 3D-automated breast ultrasound in mammographically dense breasts, Clinical Imaging 37(3) (2013), 480–486.

Gudigar

, Raghavendra

, Devasia

. et al., Global weighted LBP based entropy features for the assessment of pulmonary hypertension, Pattern Recognition Letters 125 (2019), 35–41.

Irfan

, Almazroi

A.A.

, Rauf

H.T.

. et al., Dilated Semantic Segmentation for Breast Ultrasonic Lesion Detection Using Parallel Feature Fusion, Diagnostics 11(7) (2021), 1212.

Kadry

, Rajinikanth

, Koo

, Kang

B.G.

Image multi-level-thresholding with Mayfly optimization, International Journal of Electrical & Computer Engineering 11(6) (2021).

Sreelatha

, Reddy

V.K.

Integrity and memory consumption aware electronic health record handling in cloud, Concurrent Engineering 29(3) (2021), 258–265.

Kassem

M.A.

, Hosny

K.M.

, Damaševičius

, Eltoukhy

M.M.

Machine Learning and Deep Learning Methods for Skin Lesion Classification and Diagnosis: A Systematic Review, Diagnostics 11(8) (2021), 1390.

10.

Kelly

K.M.

, Dean

, Comulada

W.S.

, Lee

S.J.

Breast cancer detection using automated whole breast ultrasound and mammography in radiographically dense breasts, European Radiology 20(3) (2010), 734–742.

11.

Khan

M.A.

, Sharif

, Akram

. et al., Skin lesion segmentation and multiclass classification using deep learning features and improved moth flame optimization, Diagnostics 11(5) (2021), 811.

12.

Kozegar

, Soryani

, Behnam

. et al., Breast cancer detection in automated 3D breast ultrasound using iso-contours and cascaded RUSBoosts, Ultrasonics 79 (2017), 68–80.

13.

, Wang

, Guo

. et al., Developing and verifying automatic detection of active pulmonary tuberculosis from multi-slice spiral CT images based on deep learning, Journal of X-Ray Science and Technology 28 (2020), 939–951.

14.

Yang

, Abubakar

A.H.

, Jiang

Deep learning and complex network theory based analysis on socialized manufacturing resources utilisations and an application case study, Concurrent Engineering 29(3) (2021), 236–248.

15.

Marr

, Hildreth

Theory of edge detection, Proceedings of the Royal Society of London. Series B. Biological Sciences 207(1167) (1980), 187–217.

16.

Thayumanavan

, Ramasamy

An efficient approach for brain tumor detection and segmentation in MR brain images using random forest classifier, Concurrent Engineering 29(3) (2021), 266–274.

17.

Mirniaharikandehei

, Heidari

, Danala

. et al., Applying a random projection algorithm to optimize machine learning model for predicting peritoneal metastasis in gastric cancer patients using CT images, Computer Methods and Programs in Biomedicine 200 (2021), 105937.

18.

Nijiati

, Zhang

, Abulizi

. et al., Deep learning assistance for tuberculosis diagnosis with chest radiography in low-resource settings, Journal of X-Ray Science and Technology 29(5) (2021), 785–796.

19.

Nugroho

H.A.

, Triyani

, Rahmawaty

, Ardiyanto

Computer Aided Diagnosis using Margin and Posterior Acoustic Featuresfor Breast Ultrasound Images, Telkomnika 4 (2017), 1776–1784.

20.

Ozmen

, Dapp

, Zapf

. et al., Comparing different ultrasound imaging methods for breast cancer detection, IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 62(4) (2015), 637–646.

21.

Rajakumar

M.P.

, Sonia

Uma B. Maheswari and S.P. Karuppiah, Tuberculosis detection in chest X-ray using Mayfly-algorithm optimized dual-deep-learning features, Journal of X-Ray Science and Technology 29(6) (2021), 961–974.

22.

Rajinikanth

, Kadry

, Nam

Convolutional-neural-network Assisted segmentation and SVM classification of brain tumor in clinical MRI slices, Information Technology and Control 50(2) (2021), 342–356.

23.

Rajinikanth

, Kadry

, Taniar

. et al., Breast cancer detection using thermal images with Marine-Predators- Algorithm selected features. In 2021 Seventh International conference on Bio Signals, Images, and Instrumentation (ICBSII) (2021), 1–6.

24.

Rajinikanth

, Sivakumar

, Hemanth

D.J.

. et al., Automated classification of retinal images into AMD/non-AMD Class—a study using multi-threshold and Gaussian-filter enhanced images, Evolutionary Intelligence 14(2) (2021), 1163–1171.

25.

Sethy

P.K.

, Pandey

, Khan

. et al., A cost-effective computer-vision based breast cancer diagnosis, Journal of Intelligent & Fuzzy Systems 41(5) (2021), 5253–5263.

26.

Shi

, Cheng

H.D.

, Hu

et al., Detection and classification of masses in breast ultrasound images, Digital Signal Processing 20(3) (2010), 824–836.

27.

Singh

B.K.

, Verma

, Thoke

A.S.

Adaptive gradient descent backpropagation for classification of breast tumors in ultrasound imaging, Procedia Computer Science 46 (2015), 1601–1609.

28.

Sood

, Rositch

A.F.

, Shakoor

et al., Ultrasound for breast cancer detection globally: a systematic review and meta-analysis, Journal of Global Oncology 5 (2019), 1–17.

29.

Taghiyev

, Altun

A.A.

, Caglar

A hybrid approach based on machine learning to identify the causes of obesity, Journal of Control Engineering and Applied Informatics 22(2) (2020), 56–66.

30.

Vijaykumar

, Baskaran

, Gayathri

Design of 4×4 antenna array for breast cancer detection, Analog Integrated Circuits and Signal Processing 105(3) (2020), 395–406.

31.

Vijayakumar

, Kadam

V.J.

, Sharma

S.K.

Breast cancer diagnosis using multiple activation deep neural network, Concurrent Engineering 29(3) (2021), 275–284.

32.

Wang

, Zhang

, Chen

. et al., Challenges in the diagnosis of testicular infarction in the presence of prolonged epididymitis: Three cases report and literature review, Journal of X-Ray Science and Technology 28(4) (2020), 809–819.

33.

Wang

, Chen

, Yang

. et al., Classification of mice hepatic granuloma microscopic images based on a deep convolutional neural network, Applied Soft Computing 74 (2019) 40–50.

34.

WHO, 2018. https://www.who.int/news-room/fact-sheets/detail/breast-cancer

35.

Williamson

, Vijayakumar

Artificial intelligence techniques for industrial automation and smart systems, Concurrent Engineering 29(3) (2021), 291–292.

36.

Zervoudakis

, Tsafarakis

A mayfly optimization algorithm, Computers & Industrial Engineering 145 (2020), 106559.