A Machine Learning Approach to Predict Blood Cancer from Patients' Symptoms and Blood Images

Abstract

This article presents a novel investigation into the detection of blood cancer and its subtypes using machine learning (ML) algorithms, with a specific focus on patient symptoms and microscopic blood images. Early diagnosis of blood cancer poses significant challenges and potential life-threatening consequences, primarily due to the complexities involved in identifying subtypes prevalent among both adults and children. This study delves into the rapid advancements in ML technologies and their applications in cancer research. Over the past decade, ML algorithms have demonstrated considerable promise in early cancer testing and have seen increased adoption in cancer diagnosis. In this research, we utilize a comprehensive ensemble of ML classifiers, including Naive Bayes, K-Nearest Neighbor, Support Vector Machine, Decision Tree, and Random Forest, to classify patient symptoms. Furthermore, state-of-the-art classifiers such as CNN, VGG, Inception, KNN, SVM, Random Forest, and Naïve Bayes distinguish between healthy and unhealthy cells in microscopic blood images. Our study's results showcase the highest accuracy achieved for the symptoms determination model using the Random Forest classifier at 85%. Additionally, microscopic image analysis demonstrates an accuracy of 97% using the Efficient Net b3 detection model. The novelty of this research lies in its pioneering approach to disease detection by integrating advanced ML algorithms to analyze both patient symptoms and microscopic images. By offering valuable insights and achieving a substantial step forward in the quest to combat blood cancer effectively, these findings contribute to the field of medical research. The integration of machine learning into cancer detection has the potential to transform the medical landscape, providing better patient outcomes and improving overall healthcare practices.