Redefining Bone Marrow Diagnostics With AI-Driven EfficientNetB5 Technology
Anurag Bhashkar, Kanwarpartap Singh Gill, G Sunil · 2024
Classifying bone marrow images into different categories is challenging because there are problems like uneven class distribution, large differences within the same class, and the need for clear and accurate diagnostic models. These challenges make it hard to create reliable automated systems for clinical diagnosis. To find different blood diseases, it's important to correctly identify images of bone marrow. There are several challenges in multi-class classification, like needing clear and correct results, a lot of difference in shape among the classes, and imbalanced classes. This work looks at using a modern convolutional neural network called EfficientNetB5 to help classify bone marrow images. Our method gets 96% accuracy by using special loss functions, new ways to improve data, and easy-to-understand tools. The results show that EfficientNetB5 can make diagnoses more reliable and accurate in medical settings. We used special loss functions to fix class imbalance, advanced data enhancement methods to make our dataset more varied, and explainable techniques to ensure our model's predictions are clear and reliable, in order to address these challenges. By using EfficientNetB5's strong design, our approach gets an impressive 96% accuracy in classification, which is a big improvement.