Utilizing Fine-Tuning ResNet-18 for Acute Leukemia Diagnosis from Blood Smear Images
Triandes Sinaga, Ade Candra, Bedy Purnama · 2024
This study explores fine-tuning ResNet-18 to enhance the accuracy of acute leukemia diagnosis from blood smear images. Early detection is crucial for patient outcomes, but the application of ResNet-18 in this context has been underutilized. We propose fine-tuning ResNet-18 to increase diagnostic precision using datasets from RSUP Haji Adam Malik Medan and Università degli Studi di Milano Statale. The model achieved 99.12% accuracy on validation and test datasets, with excellent precision, recall, F1-score, and AUCROC. Comparative analysis with non-fine-tuned ResNet-18, VGG-16, and MobileNetV2 demonstrated the superiority and stability of fine-tuning, highlighting its importance in improving diagnostic model reliability.