Performance Validation

An extensive set of simulation analysis takes place for investigating the goodness of the projected SIFT-SVM model. The results are examined under diverse dimensions, which are discussed below.

Data Set Used

To validate the results of the projected SIFT-SVM model, a benchmark skin lesion data set is employed. The information related to the data set is listed out in Table 12.1 and a few test images are demonstrated in Fig. 12.3.

TABLE 12.1

Data Set Description

FIGURE 12.3 Sample Test Images.

Results Analysis

A clear representation of the results attained by the SIFT-SVM model is shown in Fig. 12.4. The sensitivity analysis of the SIFT-SVM with compared models is shown in Fig. 12.5. The values in Table 12.2 indicate that the Li et al. method has offered the least classifier results by offering the least sensitivity value of 82%. Then, the Yuan et al. model has provided slightly better classifier results by offering

FIGURE 12.4 (a) Original image, (b) preprocessed image, (c) masked image, (d) segmented image, (e) classified image.

FIGURE 12.5 Comparative results analysis in terms of sensitivity.

a sensitivity of 82.50%. Next, the technique devised by Al-Magni et al. leads to a moderate sensitivity value of 85.40%. Afterward, the ResNet-50 and Halil et al. models have resulted in manageable and near-identical results of 90% and 90.82% respectively. Following, the VGG-19 model has resulted in a near-optimal sensitivity value of 95%. But the SIFT-SVM model performs well and achieved a higher sensitivity value of 96.23%.

The specificity analysis of the SIFT-SVM with compared models is shown in Fig. 12.6. The values in Table 12.2 indicate that the ResNet-50 method has offered the least classifier results by offering the least specificity value of 61%. Then, the VGG-19 model has provided slightly better classifier results by offering a specificity of 68%. Next, the technique devised by Halil et al. leads to a moderate specificity value of 92.68%. Following, the Al-Magni et al. model has resulted in a near- optimal specificity value of 96.69%. Afterward, the Yuan et al. and Li et al. models have resulted to a manageable and near-identical results of 97.50% and 97.80%,

TABLE 12.2

Comparison of Proposed Method with State of the Art Methods

FIGURE 12.6 Comparative results analysis in terms of specificity.

respectively But the SIFT-SVM model performs well and achieved a higher specificity value of 98.16%.

The accuracy analysis of the SIFT-SVM with compared models is shown in Fig. 12.7. The values in Table 12.2 indicate that the ResNet-50 method has offered the least classifier results by offering the least accuracy value of 75.50%. Then, the VGG-19 model has provided slightly better classifier results by offering an accuracy of 81.20%. Next, the technique devised by Li et al. leads to a moderate accuracy value of 93.20%. Afterward, the Halil et al. and Yuan et al. models have resulted in manageable and near-identical results of 93.39% and 93.40%, respectively. Following, the

FIGURE 12.7 Comparative results analysis in terms of accuracy.

Al-Magni et al. model has resulted in a near-optimal accuracy value of 94.03%. But the SIFT-SVM model performs well and achieved a higher accuracy value of 95.42%.

Conclusion

This chapter has introduced a new SIFT-SVM model to detect and classify the skin lesion from dermoscopic images. The SIFT-SVM operates using a set of subprocesses namely preprocessing. SIFT-based feature extraction, К-means clustering based segmentation, and SVM-based classification. The validation of the SIFT-SVM takes place using a skin image data set and the obtained outcome pointed out the effectiveness of the applied model over the compared methods. The obtained outcome depicted that the projected SIFT-SVM has resulted in effective results with a sensitivity of 96.23%, a specificity of 98.16%, and an accuracy of 95.42%, respectively. These values verified the enhanced outcome of the proposed model. As a part of future work, the SIFT-SVM model can be improved by the use of deep learning techniques.

References

[1] Elmisery. A. M.. Rho, S„ and Botvich, D., A fog based middleware for automated compliance with OECD privacy principles in internet of healthcare things, IEEE Access, 4, 8418-8441,2016.

[2] Zhang, C.. Lai, C. F„ Lai, Y. H.. Wu. Z. W. and Chao. H. C. An inferential real-time falling posture reconstruction for Internet of healthcare things. Journal of Network and Computer Applications, 89, 86-95, 2017.

[3] Feng, J.. Isern, N. G.. Burton. S. D., and Hu, J. Z. Studies of secondary melanoma on C57BL/6J mouse liver using 1H NMR metabolomics. Metabolites, 3, 1011-1035. 2013.

[4] Jemal, A.. Siegel. R., Ward, E., Hao. Y., Xu, J.. and Thun. M. J. Cancer statistics. 2019. CA: A Cancer Journal for Clinicians, 69, 7-34, 2019.

[5] Tarver, T. American Cancer Society. Cancer facts and figures 2014, Journal of Consumer Health Internet, 16. 366-367, 2012.

[6] Siegel, R.. Miller. K., and Jemal, A. Cancer statistics, 2018. CA: A Cancer Journal for Clinicians, 68, 7-30, 2017.

[7] Pellacani, G.. and Seidenari, S. Comparison between morphological parameters in pigmented skin lesion images acquired by means of epiluminescence surface microscopy and polarized-light videomicroscopy. Clinical Dermatology, 20, 222-227, 2002.

[8] Ali, A.-R. A., and Deserno. T. M. A systematic review of automated melanoma detection in dermatoscopic images and its ground truth data, in Medical Imaging 2012: Image Perception, Observer Performance, and Technology Assessment: International Society for Optics and Photonics: Bellingham. WA. USA. 2012, 8318.

[9] Sinz. C.. Tschandl. P, Rosendahl, C„ Akay, B. N„ Argenziano, G., Blum. A.. Braun. R. P. Cabo, H., Gourhant, J.-Y., Kreusch, J., et al. Accuracy of dermatoscopy for the diagnosis of nonpigmented cancers of the skin. Journal of American Academic Dermatology, 77, 1100-1109, 2017.

[10] Bi. L., Kim, J.. Ahn, E.. Kumar, A., Fulham. M.. and Feng. D. Dermoscopic image segmentation via multi-stage fully convolutional networks, IEEE Transactions of Biomedical Engineering, 64, 2065-2074, 2017.

[11] Schaefer, G.. Krawczyk. B., Celebi, M. E.. and Iyatomi, H. An ensemble classification approach for melanoma diagnosis, Memetic Computing, 6, 233-240, 2014.

[12] Xie, F., and Bovik, A. C. Automatic segmentation of dermoscopy images using selfgenerating neural networks seeded by genetic algorithm, Pattern Recognition. 46, 1012-1019. 2013.

[13] Shankar, K., Lakshmanaprabu, S. K., Gupta. D.. Maseleno. A., and De Albuquerque, V. H. C. Optimal feature-based multi-kernel SVM approach for thyroid disease classification, Journal of Supercomputing. 1-16, 2018.

[14] Lakshmanaprabu. S. K., Mohanty, S. N.. Shankar. K.. Arunkumar, N„ and Ramirez. G. Optimal deep learning model for classification of lung cancer on CT images. Future Generation Computer Systems. 92, 374-382, 2019.

[15] Shankar, K., Elhoseny. M., Lakshmanaprabu. S. K., Ilayaraja, M., Vidhyavathi. R. M., Elsoud. M. A., and Alkhambashi, M. Optimal feature level fusion based ANFIS classifier for brain MRI image classification. Concurrency and Computation: Practice and Experience, 32 (1). 1—12, 2018.

[16] Elhoseny. M., and Shankar, K. Optimal bilateral filter and convolutional neural network based denoising method of medical image measurements. Measurement. 143, 125-135, 2019.

13 An loHT-Based Image