A Transfer-Learning Based Unbiased Voting Bone Cancer Detection Framework from Histological Osteosarcoma Images

The most lethal types of bone cancer, osteosarcoma, have a substantially higher fatality rate for the child. The clinician’s labor can be significantly decreased and patient outcomes may be enhanced by using computer vision-based technology. A paradigm for the hard-voting, impartial identification of bone cancer is presented in our study. A histological osteosarcoma image dataset with three separate groups that is publicly available is employed in this research. A comparison of the models developed from balanced and unbalanced training sets has been accomplished using seven customized deep transfer learning algorithms, including VGG19, MobileNetV1, MobileNetV2, ResNetV250, InceptionV2, NasNetMobile, and EfficientNetV2-B0. Kappa scores and accuracy are foremost for MobileNetV2 and those are 87.55% and 92.14% respectively. In contrast, the creative Max Voting classifier eventually integrates a number of weakly performing transfer learning classifiers and exhibits robust performance on malignant categories. The Max voting classifier outperforms other existing models. The accuracy rate of the Max Voting classifier is 93.88%.