Melanoma Prediction Using CNN Methods. Part 1

Abstract Part 2

Motivation

Skin care is a critical daily topic for many people around the World. The web portal (aad.org, stats-melanoma, 2022) informs that “Melanoma can strike anyone. In fact, more than 1 million Americans are living with melanoma”. People care about their skin to avoid the developing of worst scenarios such as melanoma. Melanoma of the skin or skin cancer is in a group of highest survival in USA, the percentage is 93% (Giaquinto, A.N. et al., 2022). At the same time, regarding (Giaquinto, A.N. et al., 2022), “historically difficult-to-treat cancers, such as metastatic melanoma”. However, the paper (Giaquinto, A.N. et al., 2022) mentioned that during the recent period there is a valuable progress in metastatic melanoma treatment.

• The medicine expertise to identify whether mole has melanoma or not is difficult task. The paper (Kousis, I. et al., 2022) mentioned that “experienced specialist dermatologists had a success rate of only 60% until the invention of dermοscopic images, which increased success to between 75% and 84%”. Medical specialist cannot find all cases of melanoma. That is why, machine learning methods can used for help in melanoma prediction. Melanoma identification is visual procedure, it means, we use images of moles for finding any symptoms of skin cancer. Convolutional Neural Networks (CNN) architecture are used mostly with images and video input data. Also, there were tests to compare CNN methods prediction of melanoma results to the results of dermatologists. The paper (Esteva, A. et al., 2017) published the results of the tests: “The CNN achieves performance on par with all tested experts across both tasks, demonstrating an artificial intelligence capable of classifying skin cancer with a level of competence comparable to dermatologists. The CNN achieves 55.4 ± 1.7% overall accuracy whereas the same two dermatologists attain 53.3% and 55.0% accuracy”. With reference to the paper (Das, K. et al., 2021) “CNN has significant potential to assist dermatologists in the accurate diagnosis of melanoma”. CNN methods have similar results of melanoma prediction in comparison with dermatologists (Das, K. et al., 2021).

Table 1 CNN methods vs dermatologists regarding the paper (Esteva, A. et al., 2017)

At the same time, it is important to mention, that regarding (Al-Bander, B. et al., 2021), “Dermatologists’ manual review is also very tedious, time-consuming, subjective, and error-prone. In the US, South Australia, and Europe, the ratio of dermatologists per 1.0 million population is 34.0, 26.0, and 59.34, respectively, which are very few compared to the needed numbers.”. With reference to that, melanoma prediction mobile applications can help people around the world in their moles tracing and medical specialists can set the diagnose with higher percentage.

Table 2 The number of dermatologists per 1 million population (Al-Bander, B. et al, 2021)

• Developing CAD system for melanoma prediction requires CNN architectures, data augmentation and training setup for finding better accuracy. Besides, there are melanoma classification challenges, where researchers provide their CNN solutions. These challenges help to improve existed CNN models and discover new ones. My product uses CAD technique (Doi, K., 2007) for melanoma prediction. One of the CAD techniques is deep learning. In my system, I use CNN architectures as CAD technique for melanoma prediction based on skin lesions datasets with normal moles and moles with melanoma.
• The paper (Esteva, A. et al., 2017) presents the results of experiments for melanoma prediction of two dermatologists and CNN methods. With reference to the results in (Esteva, A. et al., 2017), when CNN method has approximately 69% for three-way accuracy, the dermatologist 1 has 65.6% of three-way accuracy and the dermatologist 2 has 66% of three-way accuracy. At the same time, when the task has 9 classes, regarding the paper (Esteva, A. et al., 2017), the dermatologists provide better results. Table 3 contains these data from the paper (Esteva, A. et al., 2017). My work uses 2 classes: melanoma and not melanoma. That is why, I expect that my machine learning product can be used by users, when people do not have any opportunities to visit dermatologists.

Table 3 CNN methods vs dermatologists’ classes 3 and 9 (Esteva, A. et al., 2017)

When we increase the number of classes, then dermatologists demonstrate better results than CNN methods. It means, that binary classification is better applied for machine learning than multi classification in comparison with dermatologists.

References

aad.org, stats-melanoma, 2022, Melanoma Available at: https://www.aad.org/media/stats-melanoma.

[Accessed 30 May 2022]

Giaquinto, A.N., Miller, K.D., Tossas, K.Y., Winn, R.A., Jemal, A. and Siegel, R.L., 2022. Cancer statistics for African American/Black People 2022. CA: A Cancer Journal for Clinicians, 72(3), pp.202–229.

[Accessed 30 May 2022]

Esteva, A., Kuprel, B., Novoa, R.A., Ko, J., Swetter, S.M., Blau, H.M. and Thrun, S., 2017. Dermatologist-level classification of skin cancer with deep neural networks. nature, 542(7639), pp.115–118.

[Accessed 30 May 2022]

Doi, K., 2007. Computer-aided diagnosis in medical imaging: historical review, current status and future potential. Computerized medical imaging and graphics, 31(4–5), pp.198–211.

[Accessed 05 July 2022]

Kousis, I., Perikos, I., Hatzilygeroudis, I. and Virvou, M., 2022. Deep Learning Methods for Accurate Skin Cancer Recognition and Mobile Application. Electronics, 11(9), p.1294.

[Accessed 29 May 2022]

Al-Bander, B., Yas, Q.M., Mahdi, H. and Al-Hamd, R.K.S., 2021. Benchmarking of deep learning algorithms for skin cancer detection based on ahybrid framework of entropy and VIKOR techniques. Turkish Journal of Electrical Engineering and Computer Sciences, 29(8), pp.2634–2648.

[Accessed 05 July 2022]

Das, K., Cockerell, C.J., Patil, A., Pietkiewicz, P., Giulini, M., Grabbe, S. and Goldust, M., 2021. Machine learning and its application in skin Cancer. International Journal of Environmental Research and Public Health, 18(24), p.13409.

[Accessed 29 May 2022]