Navigating the Challenges of Multi-Class Bone Marrow Classification with EfficientNetB5 CNN Model
Ghanshyam Agrawal, Kanwarpartap Singh Gill, Nitin Thapliyal, Ramesh Singh Rawat · 2024
Classifying bone marrow images into different groups is difficult because there are issues like unequal amounts of samples in each group, a lot of differences within the same group, and the need for very accurate and understandable diagnostic models. These challenges make it hard to create reliable automated systems for diagnosing medical conditions. To identify different blood diseases, it’s important to accurately classify bone marrow samples. There are a few challenges to deal with in multi-class classification. These include needing clear and accurate results, dealing with a lot of different shapes and sizes, and having an unequal number of examples for each class. This study looks at how a new type of computer program called EfficientNetB5 can help classify images of bone marrow. Our method gets 96% accuracy by using special loss functions, creative ways to enhance data, and techniques to understand the results better. The results show that EfficientNetB5 can make medical diagnoses more reliable and accurate. We used special loss functions to fix class imbalance, advanced data enhancement methods to make our dataset more varied, and techniques to make sure our model’s predictions are clear and reliable to tackle these challenges. Using the strong design of EfficientNetB5, our approach reaches an impressive 96% accuracy in classification, which is a big improvement.