Ön-Eğitimli Derin Öğrenme Modelleriyle Periferik Kan Hücresi Görüntülerinin Sınıflandırılması: Bir Transfer Öğrenme Yaklaşımı

Accurate and rapid classification of peripheral blood cells is critically important in the diagnosis of hematological diseases. Traditional methods involve microscopic examination by specialist physicians, which is time-consuming and subject to inter-observer variability. This study presents a deep learning-based approach for the automated classification of peripheral blood cells. The study utilizes a Peripheral Blood Cell (PBC) dataset obtained from peripheral blood smears, encompassing eight different cell types: neutrophils, eosinophils, basophils, lymphocytes, monocytes, platelets, immature granulocytes, and erythroblasts – clinically significant cell classes. The image classification problem was addressed using a transfer learning method based on deep convolutional neural networks pre-trained on large-scale datasets such as ImageNet, and the classification performance of different architectures was comparatively examined. During the training process, only the final classifier layers were retrained to reduce computational costs and ensure more efficient learning with limited data. In addition, the generalization ability of the model was improved and the risk of overfitting to the training data was reduced by applying online data enhancement techniques such as rotation, horizontal/vertical flipping, and brightness adjustment. The performance of the models was comprehensively evaluated using accuracy, precision, sensitivity, F1 score, and ROC-AUC metrics. The findings show that deep and optimized architectures achieve higher success, especially in distinguishing morphologically similar cell types. The study results reveal that the developed approach has the potential to provide experts with a fast, objective, and reliable decision support system in hematological diagnostic processes; thus, it can contribute to accelerating clinical workflows and improving diagnostic consistency.