Optimized Convolutional Neural Networks for Enhanced Detection of Acute Lymphoblastic Leukemia from Grayscale Cell Images

Acute Lymphoblastic Leukemia (ALL) is a highly aggressive blood cancer that predominantly affects young children and the elderly, with significantly varied cure rates. Traditional diagnostic methods, such as the Complete Blood Count (CBC) and peripheral blood smear, while effective, are increasingly complemented by advanced machine learning techniques for enhanced accuracy in diagnosis. This study explores the application of Convolutional Neural Networks (CNNs) to improve the prediction accuracy of ALL diagnoses by systematically tuning various parameters of the CNN model. Using a dataset from Kaggle, which includes grayscale images of cancer cells, this employed a data preparation pipeline that involved image resizing, normalization, and feature extraction through Histogram of Oriented Gradients (HOG), followed by dimensionality reduction with Principal Component Analysis (PCA). The CNN model was trained using TensorFlow and Keras, focusing on optimizing key hyperparameters such as the number of epochs, batch size, and loss functions. The findings demonstrate that a configuration using 15 epochs, a batch size of 64, and the definite cross-entropy loss function achieves the highest accuracy and efficiency in classifying leukemia images. This research not only contributes to the enhancement of leukemia detection technology but also provides valuable insights into optimizing deep learning models for broader medical applications.