Blood Cell Cancer Classification Using ResNet50 and Transfer Learning: Enhancing Diagnostic Accuracy in Hematology

The study proposes the use of a deep learning-based method, ResN et50 with transfer learning for the automatic classification of cancers in the blood cells. Blood cell cancers are one of the major health burdens around the world; thus, their detection has to be done as early and accurately as possible. Conventional microscopic examination is time-consuming with probable errors; hence, this has motivated the need for automation. In this work, ResNet50 was fine-tuned to make a robust classification among 10,868 images of blood cells, divided into six classes: eosinophil, platelet, erythroblast, monocyte, basophil, and lymphocyte. Data preprocessing and augmentation techniques were performed to improve generalization and avoid overfitting problems. The model showed high precision and recall for all classes, especially with high overall accuracy at 94%, with excellent classification, especially for eosinophils and platelets. This model had very minimal misclassifications, was very strong in generalization, and with very minimal signs of overfitting. This is an indication that deep learning can change the current diagnosis of blood cancer by offering a more rapid and accurate alternative to the manual methods. Future research endeavored to integrate this system into real-time clinical setups should, therefore, be encouraged to further elevate early detection of cancer and generally improve patient outcomes.