Human Skin Disease Detection Using Machine Learning

Abstract

In this research, I want to improve deep convolution neural networks that have been successful with the ImageNet dataset at categorizing seven different types of skin lesions using the HAM10000 dataset, which has 10,000 dermatoscopic pictures. With VGG16, Inception V3, Inception ResNet V2, and Dense Net 201, the top layers were fine-tuned.

Introduction

I. INTRODUCTION

The several processing layers are used in the deep learning approach to teach data representation in hierarchies. With just a few people working on it, it provides a method to harness a lot of data. Beginning with the development of AlexNet in 2012, the Deep Learning approach has made tremendous strides and evolution in computer vision in recent years.Identifying differences between photographs of various entities is a fairly generic skill that may be used to a variety of challenges.              

Since the very final layers of the network learn the semantics and high-level features, Deep CNN has the unique property that its initial levels often learn highly generic and "low-level" properties of pictures.

The following are the project's works:

1. Fine-tune DCNNs for 10000 dermoscopic images of 7 different types of skin lesions.
2. Inception V3, Dense Net 201, is used to fine-tune all of the layers.
3. Evaluate the performance of the following DCNNs: Dense Net 201, Inception ResNet V2, and VGG16. Every DCNN is adjusted from the top layers down.
4. Construct a seamless ensemble of Inception V3  and Dense Net 201.

II. PROBLEM STATEMENT

The purpose of our system is to make predictions for the general and more commonly occurring disorder that when unchecked can become fatal diseases. The system applies data mining techniques, does pre-processing on the data and then implements the Deep Learning algorithms.

This system will forecast the prospective ailment based on the symptoms provided and the preventative steps needed to prevent the condition from getting worse. It will also help doctors study the trends of currently prevalent diseases.

III. RESEARCH OBJECTIVE

The goal of this project is to forecast diseases in advance in order to save lives, lower treatment costs, and prevent diseases from developing in the first place.

The non-manual medical method, which is excellent for enhancing and comprehending human health, should be adopted in India as well.

The major goal is to improve patient care by applying the theory of machine learning to the healthcare industry.

Various diseases may now be identified and predicted considerably more easily because to machine learning. Numerous machine learning algorithms are used in predictive disease analysis, which aids in both disease prediction and patient treatment.

IV. RESEARCH CHALLENGE

  Infrastructure Requirements for Testing & Experimentation

1. Time-Consuming Implementation
2. Affordability
3. Clutter in the Background
4. Requires large dataset

The outcomes of fine-tuning all layers are superior to those of fine-tuning simply the top layers, and it takes less time to do so as well. This is due to the fact that I only performed for 20 epochs when fine-tuning all layers, whereas I perform for 30 epochs when fine-tuning the top layers.The results wouldn't be as good if I merely fine-tuned the top layers for a few epochs. According to this finding, fine-tuning the entire model leads to better final results and speeds up convergence of the model compared to just the top layers.Dense Net 201 provides the best single outcome in both scenarios, which is amazing considering that this model has even less parameters than Inception V3. thick Net is a very thick, deep model with few parameters, as stated in [5]. The effectiveness of DenseNet 201 in this experiment confirms the validity of employing DenseNet 201 for transfer learning on a dataset from a totally new domain that was pre trained on ImageNet. I used ensemble learning to generate an ensemble of the previously fully-tuned Inception V3 and DenseNet 201 models,and I got the best results with 88.8% accuracy on the validation set and 88.52% accuracy on the test set.

Conclusion

By using the techniques of transfer learning and ensemble learning, I was able to assemble a fine-tuned version of Inception V3 with DenseNet 201 that produced accuracy for HAM10000 of 88.52% on the test set and 88.8% on the validation set. Through testing, I\'ve discovered that fine-tuning the entire model for this dataset not only produces better results overall but also hastens the model\'s convergence. Overfitting is one severe issue that has been noted throughout training. All of my experiments have a 10–13% overfit to the training set. There are numerous techniques to reduce overfitting, but I was unable to reduce it any further.The models will improve when future efforts are made to avoid overfitting and develop better training strategies.

References

[1] Acharya, U. R., Fujita, H., Oh, S. L., Hagiwara, Y., Tan, J. H., & Adam, M. (2017). Application of deep convolutional neural network for automated detection of myocardial infarction using ECG signals.Information Sciences 415–416, 190–198. https://doi.org/10.1016/j.ins.2017.06.027 [2] Ahmed, S., Choi, K. Y., Lee, J. J., Kim, B. C., Kwon, G. R., Lee, K. H., & Jung, H. Y. (2019). Ensembles of Patch-Based Classifiers for Diagnosis of Alzheimer Diseases. IEEE Access, 7, 73373–73383. https://doi.org/10.1109/ACCESS.2019.2920011 [3] Naqi, S. M., Sharif, M., & Jaffar, A. (2020). Lung nodule detection and classification based on geometric fit in parametric form and deep learning. Neural Computing and Applications, 32(9), 4629–4647. https://doi.org/10.1007/s00521-018-3773-x [4] Rustam, F., Reshi, A. A., Mehmood, A., Ullah, S., On, B. W., Aslam, W., & Choi, G. S. (2020). COVID-19 Future Forecasting Using Supervised Machine Learning Models. IEEE Access, 8, 101489–101499. https://doi.org/10.1109/ACCESS.2020.2997311 [5] Liu, J., Xu, H., Chen, Q., Zhang, T., Sheng, W., Huang, Q., Song, J., Huang, D., Lan, L., Li, Y., Chen, W., & Yang, Y. (2019). Prediction of hematoma expansion in spontaneous intracerebral hemorrhage using support vector machine. EBioMedicine, 43, 454–459. https://doi.org/10.1016/j.ebiom.2019.04.040 [6] Javeed, A., Zhou, S., Yongjian, L., Qasim, I., Noor, A., & Nour, R. (2019). An Intelligent Learning System Based on Random Search Algorithm and Optimized Random Forest Model for Improved Heart Disease Detection. IEEE Access, 7, 180235–180243. https://doi.org/10.1109/ACCESS.2019.2952107 [7] Acharya, U. R., Fujita, H., Oh, S. L., Hagiwara, Y., Tan, J. H., & Adam, M. (2017). Application of deep convolutional neural network for automated detection of myocardial infarction using ECG signals. Information Sciences, 415–416, 190–198. https://doi.org/10.1016/j.ins.2017.06.027 [8] Kousarrizi, M. R. N., Seiti, F., & Teshnehlab, M. (2012). An Experimental Comparative Study on Thyroid Disease Diagnosis Based on Feature Subset Selection and classification. International Journal of Electrical &Computer [9] Cinarer, G., & Emiroglu, B. G. (2019). Classification of Brain Tumors by Machine Learning Algorithms. 3rd International Symposium on Multidisciplinary Studies and Innovative Technologies, ISMSIT 2019 - Proceedings. https://doi.org/10.1109/ISMSIT.2019.8932878 [10] Senturk, Z. K., & Kara, R. (2014). Breast Cancer Diagnosis Via Data Mining: Performance Analysis of Seven Different Algorithms. Computer Science & Engineering: An International Journal, 4(1), 35–46. https://doi.org/10.5121/cseij.2014.4104

Copyright

Copyright © 2024 Nandimandalam Venkata Siva Greeshmanth. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.