Evaluation of Machine Learning Algorithms in the Healthcare Sector

Introduction

Healthcare has always been a crucial business that offers care based on value to many people all over the world. This industry is growing very fast and has turned into a prime revenue earner for several countries. In the United States, the revenue from healthcare is approximately $1,668 trillion. The United States spends more on healthcare than most other countries in the world. Value, Outcome, and Quality are three keywords that are always linked to healthcare, these are promises that most specialists and stakeholders in the healthcare industry are looking to live up to.

With the rise of more and more applications using machine learning in the healthcare field, we are able to see a future where innovation and data analysis work together. We can expect the future healthcare systems all over the world to be a combination of Machine Learning based application with real-time data. This could increase the effectiveness of treatment options which are not available today. One of the most commonly used Machine Learning applications in the healthcare field is to identify and diagnose diseases that are hard to discover otherwise. This includes diseases such as cancer, which are difficult to identify in their initial stage.

Major Section

Mrunmayi Patil, Vivian Brian Lobo, Pranav Puranik, Aditi Pawaskar, Adarsh Pai et al (2018) have used support vector machine to understand and predict lifestyle diseases that an individual might be susceptible to. The have simulated an economic machine learning model as an alternative to deoxyribonucleic acid testing that analyzes an individual’s lifestyle to identify possible threats that form the foundation of diagnostic tests and disease prevention, which may arise due to unhealthy diets and excessive energy intake, physical dormancy, etc. After research, they found that if a parent has a particular disorder, it does not necessarily mean that a child would develop the same. However, there could be a possibility of high risk of developing the disorder (i.e., genetic susceptibility), and for such a possibility where it cannot be a sure occurrence, but risk prevails

Naresh Khuriwal and Nidhi Mishra (2018) have used adaptive ensemble voting method for diagnosed breast cancer. They have used Logistic Regression and Neural Networks to analyze their data. After conducting multiple experiments, the authors found that the artificial neural network approach with logistic algorithm achieved 98.50% accuracy, much higher compared to the other machine learning algorithms (Khuriwal, 2018).

This research closely related to a study conducted by Madhuri Gupta and Bharat Gupta (2018) aims to present machine learning techniques in cancer disease by applying learning algorithms on breast cancer. They have made the use of Linear regression, Random Forest, Multi-layer Perceptron and Decision Tree algorithms to analyze the data. After conducting the experiment, the authors found that performance in terms of accuracy, MLP is better as compared to other techniques. MLP technique also performs better than other techniques when Cross Validation metrics is used in breast cancer prediction

One major similarity between the two above papers is that they have only considered Wisconsin data. Had the authors picked up data which was not centralized to one specific location the results may vary.  The results may even show a completely different picture if dataset were from other parts of the world. Data could be gathered globally and compared with the results to see if this is only a local issue or it has the same effect worldwide, this could be an angle to analyze for future enhancements. However, one major difference between the two studies is that in the first study the authors used only logistic regression and neural networks, whereas the authors of the second article have evaluated many more techniques.

Another study performed by Moh’d Rasoul Al-Hadidi, Abdulsalam Alarabeyyat, and Mohannad Alhanahnah (2016) proposed a method if the patient had to detect the breast cancer or not with high accuracy. The method used consisted of two main parts.  The first part made use of the image processing techniques to prepare the mammography images for pattern and feature extraction. The extracted features are utilized as an input for a two types of supervised learning models. The algorithms used were Logistic Regression and Backpropagation Neural Network. After research, they obtained a good regression value using Backpropagation Neural Network that exceeded 93% with only 240 features (Al-Hadidi, 2016).

This research closely related to a study conducted by and Bita Shadgar (2010). They make use of support vector machines, K-nearest neighbors and probabilistic neural networks classifiers are combined with signal-to-noise ratio feature ranking, sequential forward selection-based feature selection and principal component analysis. This information is then taken to carry out feature extraction to distinguish between malignant and benign and tumors. The authors were able to achieve 98.80% and 96.33% respectively using support vector machines classifier models against two widely used breast cancer benchmark datasets (Osareh, 2010).

Yet another study on breast cancer was conducted by Arjun P. Athreya, Alan J. Gaglio, Junmei Cairns, Krishna R. Kalari, Richard M. Weinshilboum et al. (2018). They try to identify a few genes among the 23,398 genes of the human genome to establish new drug mechanisms. The authors have used different types of clustering techniques to identify the results, Hierarchical Clustering and K-Means Clustering. Based on their analysis they concluded that methods can augment the drug and disease knowledge of pharmacogenomics experts by identifying biomarkers of novel drug actions. On limitation to this experiment is that the authors have only used 2 types of clustering techniques. They could enhance their work by also implementing more clustering techniques, such as, GMM, Mean-Shift clustering and DBSCAN to see if they obtain similar results.

Another study conducted on lung cancer by Wasudeo Rahane, Himali Dalvi, Yamini Magar, Anjali Kalane, and Satyajeet Jondhale (2018). They explore efficient methods to detect lung cancer and its stages successfully. The techniques applied are Support Vector Machine and Image Processing. After analyzing the data, the authors were able to conclude that Support Vector Machine classifier, classifies the positive and negative samples of lung cancer images in the system.

Another article on lung cancer by Qing Wu and Wenbing Zhao (2017) uses a neural-network based algorithm to detect small cell lung cancer from computed tomography images. For the training set they have used CT images of ten patients, five of which have been diagnosed with small cell lung cancer and five others without it. After training the model and analyzing it, they were able to makes 10 false positive predictions and missed 6 cases when the patients actually have small cell lung cancer detection.

Though both of the above articles have conducted studies on lung cancer, the first article uses support vector machine and image processing to identify if the images are positive or negative for lung cancer. Whereas, in the second article the authors have used neural networks to do the same and have used a much larger data set.

Sangman Kim, Seungpyo Jung, Youngju Park, Jihoon Lee, and Jusung Park (2018) conducted a study to find useful markers from sensor arrays which have massive sensing points and diagnose liver cancer. The techniques they applied in their research was Neural Network and Fuzzy Neural Network. After the research was completed, they were able to detect liver cancer with the accuracy of 99.19 % by average use of 132 aptamers based on neural network and 98.19 % by average use of 226 aptamers based on fuzzy neural network (Kim, 2018). Though the accuracy achieved is very high they authors could have also used convolutional neural network to check the accuracy of their research as another comparison algorithm.

Lastly, Joseph M. De Guia, Madhavi Devaraj, and Larry A. Vea (2018), present the existing technology of microarray gene expression and classify the cancer genes using machine learning algorithms. The algorithms that they have using for this presentation are Gradient Boost and Support Vector Machine. After conducting the study, the authors for that the performance accuracy for Support Vector Machine was 58% and for Gradient was 64% (Guia, 2018). This is a very low accuracy level. The authors could try to implement other types of algorithms such as, Decision Tree and Random Forest to see if the accuracy would increase.

Conclusion

Analyzing the different types of diseases and then trying to diagnose them using machine learning is a complex task. However, using ingenious machine learning algorithms as well as medical knowledge and methods, it has been possible to identify and diagnose certain types of diseases.

Past research shows the use and results of various methods and experiments using various data sets and algorithm. For the purpose and scope of my research, I propose to mingle the methods used by other experts according to my needs and attempt to build an application that will identify and diagnose diseases in healthcare. Fuzzy Neural Network and Backpropagation Neural Network seem to be the most proficient for the purposes of this study.

Reference

• Patil M., Lobo V.B., Puranik P., Pawaskar A., Pai A., and Mishra R. (2018). A Proposed Model for Lifestyle Disease Prediction Using Support Vector Machine. 2018 9th International Conference on Computing, Communication and Networking Technologies (ICCCNT). pp. 1-6. doi: 10.1109/ICCCNT.2018.8493897
• Gupta M. and Gupta B. (2018). A Comparative Study of Breast Cancer Diagnosis Using Supervised Machine Learning Techniques. 2018 Second International Conference on Computing Methodologies and Communication (ICCMC). pp. 997-1002. doi: 10.1109/ICCMC.2018.8487537
• Wu Q. and Zhao W. (2017). Small-Cell Lung Cancer Detection Using a Supervised Machine Learning Algorithm. 2017 International Symposium on Computer Science and Intelligent Controls (ISCSIC). pp. 88-91. doi: 10.1109/ISCSIC.2017.22
• Kim S., Jung S., Park Y., Lee J., Park J., and Mishra R. (2018). Effective liver cancer diagnosis method based on machine learning algorithm. 2014 7th International Conference on Biomedical Engineering and Informatics, pp. 714-718. doi: 10.1109/BMEI.2014.7002866
• Khuriwal N. and Mishra N. (2018). Breast cancer diagnosis using adaptive voting ensemble machine learning algorithm. 2018 IEEMA Engineer Infinite Conference (eTechNxT). pp. 1-5. doi: 10.1109/ETECHNXT.2018.8385355
• Al-Hadidi M.R., Alarabeyyat A. and Alhanahnah M. (2016). Breast Cancer Detection Using K-Nearest Neighbor Machine Learning Algorithm. 2016 9th International Conference on Developments in eSystems Engineering (DeSE). pp. 35-39. doi: 10.1109/DeSE.2016.8
• Rahane W., Dalvi H., Magar Y., Kalane A. and Jondhale S. (2018). Lung Cancer Detection Using Image Processing and Machine Learning HealthCare. 2018 International Conference on Current Trends towards Converging Technologies (ICCTCT). pp. 1-5. doi: 10.1109/ICCTCT.2018.8551008
• Osareh A. and Shadgar B. (2010). Machine learning techniques to diagnose breast cancer. 5th International Symposium on Health Informatics and Bioinformatics. pp. 114-120. doi: 10.1109/HIBIT.2010.5478895
• Athreya A.P. et al.. (2018). Machine Learning Helps Identify New Drug Mechanisms in Triple-Negative Breast Cancer. IEEE Transactions on NanoBioscience. 17:3 pp. 251-259. doi: 10.1109/TNB.2018.2851997
• Guia J.M.D., Devaraj M. and Vea L.A. (2018). Cancer Classification of Gene Expression Data using Machine Learning Models. 2018 IEEE 10th International Conference on Humanoid, Nanotechnology, Information Technology,Communication and Control, Environment and Management (HNICEM). pp. 1-6. doi: 10.1109/HNICEM.2018.8666435