Optimized Dense Neural Network Architecture for Bone Marrow Cancer Classification Using Custom Callbacks

Classification of bone marrow cancer is a crucial step in the analysis of several hematological cancers, such as multiple myeloma and leukemia. This study shows better dense neural network (DNN) architecture for significant the variation among the five main types of bone marrow cells includes lymphocytes, myelocytes, band neutrophils, and plasma cells. To find and diagnose bone marrow cancers like multiple myeloma, acute myeloid leukemia (AML), and acute lymphoblastic leukemia (ALL) early on, it is important to be able to accurately identify these cell types. The suggested DNN model uses unique callbacks to optimize training, avoid overfitting, and guarantee faster convergence. These callbacks include learning rate scheduling, early pausing, and dynamic weight modifications. The system preprocesses the high-resolution bone marrow image database to enhance contrast, adapt to fluctuations in illumination, and utilize dense layers for cell classification and convolutional layers for feature extraction. After extensive hyperparameter tuning, a model that classified these five categories of bone marrow cells obtained an overall accuracy rate of 95%. In terms of accuracy, recall, and F1 scores, the model performs better than conventional machine learning techniques while retaining computing economy. The model is able to increase training times without sacrificing classification performance by incorporating custom callbacks. With its enormous clinical application potential, our improved DNN architecture offers pathologists a dependable and effective tool for accurately classifying bone marrow cells. By improving bone marrow cancer diagnosis accuracy, this approach can improve patient care and treatment results.