Osteosarcoma is the most common primary bone malignancy, predominantly affecting children and adolescents. While histopathology remains the gold standard for diagnosis and treatment decisions, machine learning and deep learning offer potential improvements. This study comparatively evaluated various deep neural networks using publicly available osteosarcoma images to assess their performance in histopathological evaluation. Results indicated that smaller networks, such as MobileNetV2, with smaller image inputs achieved the best performance (91% accuracy with 5-fold cross-validation), highlighting the importance of network and input size selection and suggesting that a larger number of parameters doesn't always equate to better performance.

Ioannis A Vezakis, George I Lambrou, George K Matsopoulos