Evaluating Performance of Machine Learning Models for Chromosomal Disorder Classification using Karyotype Images: A Comparative Study

This project presents a comprehensive study on the automated identification of chromosomal disorders using karyotype images. The goal is to enhance the accuracy and efficiency of diagnosis by leveraging machine learning techniques. The comparative analysis involves five prominent machine learning models, namely XGBoost, Random Forest, SVM, Naive Bayes, and InceptionV3. Additionally, two hybrid models are explored, combining InceptionV3 for feature extraction with either XGBoost or an Artificial Neural Network (ANN) for classification. The project tackles the challenge of limited publicly available datasets by collaborating with the Kshema Centre for Genetic Services, resulting in a dataset of 835 images. The results reveal that XGBoost and Random Forest models achieve high accuracy levels of 91% and 85%, respectively, outperforming other models. The hybrid models, combining InceptionV3 with XGBoost and ANN, also demonstrate promising performance, reaching accuracies of 78% and 60% respectively. These findings highlight the potential of hybrid models in capturing both the feature extraction capabilities of deep learning and the classification power of traditional machine learning algorithms. The successful implementation of these models paves the way for more accurate and efficient chromosomal disorder identification. By leveraging machine learning and hybrid modelling approaches, this project contributes to advancing medical image analysis and has significant implications for the field of chromosomal disorder diagnosis.