The EfficientNetB5 Solution to Multi-Class Bone Marrow Classification Challenges

The need for easy-to-understand and highly accurate diagnostic models, along with problems like uneven representation of different categories and large differences within the same category, make it difficult to classify bone marrow images into multiple groups. These tough conditions make it harder to create reliable computer systems for medical diagnosis. Bone marrow pictures need to be correctly identified to diagnose some blood-related illnesses. However, there are some big challenges in the multi-class class, like different shapes, uneven beauty, and the need for clear and personal results. This paper looks at how to use EfficientNetB5, a type of computer program that helps recognize images, to solve problems in classifying bone marrow pictures. By using advanced methods to add information, tailored loss options, and ways to understand our results, we have achieved a very high accuracy of 96%. The results suggest that EfficientNetB5 could make diagnosing problems more accurate and reliable in scientific settings. To tackle these challenges, we created special features to handle balance issues in the data, used better methods to increase the variety of our dataset, and applied techniques to make sure the model’s predictions are clear and trustworthy. We greatly improve how well we can identify categories by using the strong structure of EfficientNetB5, achieving an impressive accuracy of 96%.