Hybrid Active Contour Mammographic Mass Segmentation and Classification

This research implements a novel segmentation of mammographic mass. Three methods are proposed, namely, segmentation of mass based on iterative active contour, automatic region growing, and fully automatic mask selectionbased active contour techniques. In the first method, iterative threshold is performed for manual cropped preprocessed image, and active contour is applied thereafter. To overcome manual cropping in the second method, an automatic seed selection followed by region growing is performed. Given that the result is only a few images owing to over segmentation, the third method uses a fully automatic active contour. Results of the segmentation techniques are compared with the manual markup by experts, specifically by taking the difference in their mean values. Accordingly, the difference in the mean value of the third method is 1.0853, which indicates the closeness of the segmentation. Moreover, the proposed method is compared with the existing fuzzy C means and level set methods. The automatic mass segmentation based on active contour technique results in segmentation with high accuracy. By using adaptive neuro fuzzy inference system, classification is done and results in a sensitivity of 94.73%, accuracy of 93.93%, and Mathew’s correlation coefficient (MCC) of 0.876.


Introduction
The X-ray of breast is called a mammogram, which is used to locate cancer at an early stage. Mass is nothing but a lump that has its own shape and marginal properties. Benign is an initial stage of cancer and marginated smoothly, whereas malignancy becomes more violent with time. Mass can be categorized into three types, namely, well-defined, speculated, and ill-defined [1,2]. Numerous studies have been conducted to identify a mass in its early stages. The literature has indicated that segmentation, which is highly efficient, accurate, and fully automatic, is needed for spotting a mass at its beginning phase [3,4].

Iterative Active Contour-Based Mass Segmentation
Mammographic image with mass is considered an input and preprocessing techniques are used, which include cropping and equalization using histogram. Preprocessed images are subjected to iterative thresholding to extract a rough boundary. Active contour is applied to the iterative threshold image, leading to final segmentation. The main advantage of contour-based segmentation is that it preserves mass region information, edge, and shape. All processes are carried out and the results are obtained using MATLAB. Fig. 1 shows the architecture diagram of the iterative active contour segmentation.
Images are obtained from the MIAS database and collected from the Clarity Imaging Center, Coimbatore. The mass region is cropped and the preprocessed gray scale picture is converted into a binary image using the iterative threshold technique [5,6]. The threshold value θ is calculated iteratively based on the foreground and background strengths. The threshold value of the image shown in Fig. 2 is calculated as 128. The original, preprocessed, and threshold images are shown in Fig. 2.
Threshold image is taken as input. The mask image size is taken as the same size of the iterative threshold image. Within this image, a square mask of size 111 × 111 is defined. On the basis of energy minimization, the defined mask will fit to the boundary. Gradient descent minimization is used to minimize snake energy [7]. The force acting on the contour is as follows: The total energy of the contour is as follows:   where S v ð Þ gives the derivative of S with respect to v. The internal energy of the snake is given as follows: The external energy acting on the contour is as follows: where En image gives the force acting on the contour and En con depicts the external constraint force as follows: En line ¼ I x; y ð Þ, which depends on the sign of w line , and the width of the line varies. All edge information will be found by the edge energy function as follows: Eq. (8) represents the next iteration of the contour: The iteration repeats until the snake converges its minimal boundary [8]. Fig. 3 illustrates various stages involved in the proposed method of mass segmentation [9,10]. The output obtained (450 th iteration) for a sample image is given in the following section.
The mass affected area in the mammographic image is cropped manually and allowed for further processing. This drawback is avoided by adopting a new segmentation scheme, in which the entire image is considered without cropping.  Fig. 4 shows the flow diagram of the automatic selection of seed based on the region growing [11,12] mass segmentation technique [13,14]. As a pre-study, a mask of size 20 × 20 is initialized for various reference images. For regions within the mask, statistical features are extracted, namely, contrast, dissimilarity, sum average, sum variance, and auto correlation, as listed in Tab. 1. For the image under testing, if the extracted features lie in the range of the predefined statistical features of mass, then the seed is fixed. The input image is subjected to thresholding. Thereafter, the final gray level segmented picture is obtained. The mask is moved in case of a mismatch in the extracted features.

Automatic Mass Selection-Based Active Contour Technique
This method completely aims at automatic mask selection based on the active contour technique. The mask for the active contour technique is chosen automatically by reading the mass features. Fig. 7 illustrates the flow diagram of the fully automatic active contour-based mass segmentation.
A mask of size 20 × 20 is initialized. Five statistical features are extracted for the regions within the mask (same as in method 2). If the extracted features lie in the range of the predefined features of mass, then the mask is fixed. The input image is subjected to thresholding and active contour technique is implemented thereafter [15]. Lastly, gray level segmented image is obtained. Thereafter, the mask is moved in case of mismatch in the extracted features [16]. Fig. 8 depicts the output of the preceding method. Fig. 9 shows the output of the three proposed methods for various mammographic images. Evidently, over segmentation for image 2 is prevented by using the automatic active contour technique.
The expected output is marked manually by the physician, and the mean values of the three proposed methods are calculated in Tab. 2. Note that the difference in the mean values of the proposed method 3 is 1.085. This value is closer to zero, thereby indicating the closeness of the segmentation.
The proposed method is compared with the existing methods, such as the fuzzy C means (FCM) clustering and level set methods (see Fig. 10).
The statistical means of the preceding methods are compared using Fiji (see Tab. 3).
Tab. 3 shows that the proposed algorithm holds good because its difference of mean value is low, which shows the closeness of segmentation. ANFIS is used as a classifier [17,18]. Various membership functions and their testing errors are shown in Fig. 11. The triangular membership function with minimum error of 0.463 is considered. Three triangular membership functions are used for each input. A total of 729 rules are framed. Tab. 6 lists the number of images considered for classification.        Figure 11: Various membership functions and their testing errors The classified ANFIS plot is shown in Fig. 12. As presented in Fig. 12, a confusion matrix is formed and shown in Tab. 7.
From the confusion matrix, the following performance metrics are calculated: sensitivity, specificity, precision, accuracy, and MCC [15]: The preceding values are calculated and listed in Tab. 8.
Tab. 8 shows that accuracy is inferred to be 93.93% and MCC is 0.876. Hence, a high accurate classification is made. The proposed method serves as a second opinion for doctors in diagnosing cancer.

Conclusion
Women are mostly affected by breast cancer, thereby resulting in this disease becoming a major cause of death. Thus, technology for early detection should be identified. Evidently, the proposed method is of high medical significance in detecting early-stage cancer.