|
| Author and year | Algorithm implemented | Working | Dataset | Research outcome | Advantages |
|
| Anitha and Peter, 2015 [8] | MCSU in cellular automata | Mammographic image preprocessed to remove noise, markers, pectoral muscles, and unwanted information. Peak analysis is done and enhanced by CLAHE followed by automatic selection of seed and finally updating the cellular strength using cellular automata | Mini-MIAS, 70 samples | The specificity of the dataset is maintained and improved | The initial seed selection needs no manual interception; the automatic pick of seed point is done |
|
| Gupta and Tiwari, 2017 [14] | HM-GRA and CLAHE | The histogram of the mammographic image was generated and the selection of parameters for enhancement was performed. The modification of the histogram is done using the uniform histogram. The grey relational analysis was used to improve the contrast and further normalization and segmentation of ROI was presented | Mini-MIAS 322 samples | The contrast of the image is enhanced to improve minute calcification by decreasing the ratio of false positives | Global and local contrast is improved, and sensitivity and specificity both are taken care of at the same time |
|
| Taghanaki et al., 2017 | Geometry-based model | The maximum area in the breast contour is covered along with the boundary to be marked for early detection | INbreast, DDSM, MIAS, 197, 353, and 322 samples | The precision of ROI segmentation increased | It works perfectly with multilayered samples having a lot of variations in intensity and edge boundary |
|
| Shi et al., 2018 [16] | Gradient weight map, pixel wise clustering, and local text filter | Artifacts removed from original image further segmentation are done based on pixel-wise clustering, followed by detection of the boundary of breast muscles. Finally, a local texture filter is used to detect calcification | MIAS, BCDR, INbreast, 322, 100, and 201 samples | It is immune to noises and can detect calcification in dense breasts too. With few settings, FFDM images can also analyze effectively | The distinction of skin air boundary marked by the proposed algorithm gradient weight map in compassion to another threshold-based algorithm |
|
| Shen et al., 2018 [17] | A genetic algorithm for threshold and segmentation and morphological selection | The genetic algorithm was implemented to study multilevel threshold, segmentation, and classification based on pectoral muscle segmentation done to classify between successful, acceptable, and unacceptable | MIAS, DDSM, and INbreast | The precision of segmentation is higher in comparison to other existing algorithms | The classification between acceptable, unacceptable, and successful. The sensitivity was then checked for unacceptable samples |
|
| Hazarika and Mahanta, 2018 [18] | Pectoral muscle removal using region growing | A suppression algorithm is applied, and further, the samples whose results come comparable and close with the hand-drawn segmented mask are distinguished as accepted | Mini-mias and 150 samples | 86.67% is the accuracy of acceptable and 5.33% for partially fair | Hand-drawn segmentation mask compare the accuracy of segmentation algorithm given |
|
| Alam, et al., 2018 [19] | Segmentation using morphology | Morphological and interpolation operations are used to segment ROI, and further splitting is done based on intensity value. For creating clusters of microcalcification area, ranking is used on the differenced image | DDSM, MIAS,248, and 24 samples | The highest classification accuracy is 94.48% approximate | Dice metric similarity score was calculated to measure the evaluation, and further reference masks were also used |
|
| Anitha and Peter, 2015 [20] | KFLS (kernel-based fuzzy clustering) | One the preprocessing of the mammogram is done then ROI, which is segmented out using fuzzy C-means clustering segmentation method | DDSM and 300 samples | 94% segmentation precision in terms of sensitivity | As it is based on an intelligent system, i.e., fuzzy clustering, it provides high precision |
|
| Touil and Kalti, 2016 [21] | IFBS (iterative fuzzy breast segmentation algorithm) | The image is divided into k clusters to remove the over-segmentation background region and extract perfect ROI | MIAS and 200 samples | As compared to the manual ROI curve, its performance is 60% better | It reduces the over-segmentation of the background |
|
| Kozegar et al., 2018 [22] | DRLSE and OBNLM filter | Region growing with combination with GMM is used, followed by despeckling and fine segmentation. DRLSE was modified in the paper | Ultrasound images and 50 samples | It assumed that seed position is known before as it does not work on images with edges | Also, work where the variance is different |
|
| Aggarwal and Chatha, 2019 [23] | Edge detection algorithm is designed on 8-bit grayscale image | Binarization is done to reduce the data reduction step using an edge detection algorithm | MIAS 50 random samples | It reduced the difference between region of interest and background | Data reduction leads to loss of information can be reduced by using multilevel thresholding |
|
| Tembhurne et al., 2021 [24] | Computer-aided transfer learning-based deep model for binary classification for breast cancer detection | Multichannel merging methods for making a dual-ensemble architecture | Break-his dataset is used | Ensemble architectures by using pretrained models like Xception and InceptionV3 results in an accuracy of 97.5% is achieved | Combining different algorithms gives better accuracy over measuring accuracy from one algorithm |
|
| Malathi et.al., 2021 [25] | The algorithm uses breast CAD scheme feature fusion using CNN deep features network | The abnormality in breast images is scrutinized through deep belief network | CAD images are used | The outcome shows random forest algorithm is giving an accuracy of around 97.51% over the CNN classifier | The algorithm removes the point spread function where low-dose medical CT image restoration and recovers the reconstructed image quality, efficiency, and speed through sparse transform |
|
| Fang et al., 2021 [26] | Configuration of the multilayer perceptron (MLP) neural network multilayer perceptron network using backpropagation network | A new training algorithm is proposed based on whale optimization for MLP network | MIAS 332 digitized mammography images | Detection performance is detected using detection rate and identification with the false percentage | Accuracy is achieved as compared to other methods |
|
