EfficientNetB5 Based Classification of Bone Marrow Cells for Hematologic Disease Diagnosis
Aditya Kumar, Leema Nelson, S. Gomathi · 2024
This study presents a comprehensive study on applying deep learning techniques for bone marrow cell classification using the EfficientNetB5 model. The primary objective is to classify bone marrow cell images into seven distinct classes accurately. The dataset used in this study comprises images of bone marrow cells collected from diverse sources. Data preprocessing techniques enhanced model performance and generalisation, including data augmentation and normalisation. The proposed model architecture consists of an EfficientNetB5 base model followed by batch normalisation, dense layers, and dropout regularisation. A custom callback function was implemented to dynamically adjust learning rates during training based on model performance metrics. The training process was monitored using various metrics, such as loss, accuracy, and validation loss. The results indicate significant improvements in classification accuracy, reaching 94.66% on the test dataset. Furthermore, the model’s performance was evaluated using a confusion matrix, providing insights into classification errors and model robustness. Overall, this study demonstrates the effectiveness of deep learning approaches, particularly the EfficientNetB5 architecture, for bone marrow cell classification tasks, offering promising prospects for enhancing medical diagnosis and research in haematology.