Application of a Deep Learning Model for Effective Diagnosis of Osteosarcoma

This experiment aimed to test the effects of different osteosarcoma tumor types on the accuracy of a trained neural network. Osteosarcoma is a subset of bone cancer seen greatly in pediatric cases and young adults, and its onset involves the rapid production of abnormal bone tissue, which leads to acute injuries. It is malignant by nature and life-threatening; therefore, it must be detected early by methods including X-ray, CT, MRI, and microscopy. The diversity of neoplastic formation makes classifying difficult, so using neural networks for detection can be beneficial. These machine-learning algorithms can identify patterns and analyze medical images for an accurate diagnosis. A dataset with three osteosarcoma classes was used: Viable Tumor, Non-Viable (Necrotic) Tumor, and Non-Tumor. All images were hematoxylin and eosin (H&E) stained cells used in training and testing. The structure of the network was a pre-trained model: VGG19, which uses the structure of a Convolutional Neural Network (CNN). The structure included the input, 18 convolution layers, max-pooling layers, and hidden layers for feature extraction. The multi-class classification accuracy was 96.27%, proving to be efficient. Large amounts of patient data can be quickly deciphered, and this could lead to advancements in medical imaging and disease prognosis.