February 2022 - March 2022
Source: https://www.aaroncohen-gadol.com/patients/glioma/types-of-glioma/overview-of-types-of-glioma
Glioma is a devastating malignancy of the brain that affects thousands of people globally each year. It is the most common form of primary brain tumor and is notorious for its poor prognosis, which is largely due to inefficient tumor classification methods. Current methods rely on visual examination of tissue samples, which can be highly subjective and prone to error. This can result in delayed diagnosis and treatment, leading to poor outcomes.
To test our hypothesis, we evaluated four different supervised classification models - Gaussian Naive Bayes, K-nearest neighbor, Logistic Regression and Support Vector Machines. We used gene expression data from The Cancer Genome Atlas (TCGA) to train these models to classify between normal and glioma samples, and further classify the subtypes of glioma. We used five-fold cross-validation to evaluate our models for extracted gene expression data.
PCA results of data.
Out of these models, I was responsible for implementing Gaussian Naive Bayes from scratch, without the use of any machine learning packages. We confirmed that our codes were behaving properly by implementing a scikit version of Gaussian Naive Bayes and SVM. Additionally, we conducted a 5-fold cross-validation to validate the results.
Our comparative study of the performance of each classifier against the data showed that SVM was the best performing classifier. We believe that our results could have significant implications for the early diagnosis and treatment of gliomas, ultimately improving patient outcomes.
Final classifier evaluation
Overall, our project highlights the potential of machine learning algorithms to revolutionize the classification of gliomas and other cancers. Our findings could pave the way for more accurate and reliable tumor classification methods, leading to earlier diagnosis and more effective treatment options.
Python, sklearn
Mirudhula Mukundan, Aditi Sarathy, Ketaki Ghatole, Arnav Gupta