OSTEOSARCOMA BONE TUMOR DETECTION USING DEEP ENSEMBLE TRANSFER LEARNING ARCHITECTURE

Abstract

Being the most common bone cancer affecting children, osteosarcoma emerges during periods of rapid growth and development in bones which normally occurs during childhood and adolescence. Deriving from the under differentiated formation of osteoblasts, it leads to a haphazardly made, poorly mineralized bone matrix. Although it’s the most common bone tumor type yet the exact mechanisms controlling the onset of osteosarcoma are unknown. Only a few proteomics analyses have been per- formed so far. Consequently, all available studies have mainly identified other cancer-related proteins rather than specifically osteosarcoma-associated ones. This paper aims to emphasize the growing scope of proteomics cutting-edge methodologies, especially their customized use to develop new diagnostic and therapeutic osteosarcoma targets. Moreover, it describes recent knowledge on signaling markers and pathways related to both the original and metastatic types of osteosarcomas from early- stage proteomic profiling studies. However, there is a major class imbalance problem in this dataset. To address this problem, we used the Synthetic Minority Over- sampling Technique (SMOTE). This technique helps distribute images across classes evenly to prevent class stability issues. We evaluate the model based on precision, which stands at 98%, sensitivity or recall at 96%, and F1 score at 97%. Thus, on analysis the final outcome, we found out that the proposed CNN model performs better than the other models on all the given evaluation metrics.