Decoding Acute Lymphoblastic Leukemia: Insights from Convolutional Neural Networks and Pretrained Model
Jason Hendrawan, Jonathan Adrian, Verrel Juanto Lukmana, Felix Indra Kurniadi · 2023
Accurate detection of Acute Lymphoblastic Leukemia (ALL) is essential for prompt diagnosis and effective treatment. In this study, we introduce a deep learning method employing Convolutional Neural Networks (CNN) and pretrained models to enhance the precision of ALL detection. We leveraged a dataset comprising blood cell images for training and validation. Performance evaluations of three CNN models—ResNet-50, VGG16, and a bespoke CNN architecture—were conducted using metrics like accuracy, loss, and validation scores. Our findings reveal that VGG16 achieved a notable accuracy of 96.53%, while our custom CNN yielded 88.15%. Notably, VGG16, despite its high accuracy, exhibited signs of overfitting. In contrast, ResNet-50 underperformed, registering an accuracy of just 79.05%. This work underscores the potential of technological innovations in healthcare, especially in ALL diagnosis, and paves the way for more in-depth exploration of deep learning in cancer detection.