Osteosarcoma cancer detection using ghost-faster RCNN model from histopathological images

Subclassifying cancer tumours based on molecular characteristics holds promise for improving treatment response rates among patients. However, conventional machine learning approaches often fall short of effectively subclassifying osteosarcomas (OS) due to the intricate cellular origins involved. In this study, we address this challenge by leveraging an unequally distributed dataset comprising eosin-stained histopathological images with osteosarcoma haematoxylin. We investigate how this dataset's distribution impacts the performance and reliability of models and studies utilizing it, recognizing the complexity it introduces to osteosarcoma classification efforts. To tackle this issue, we propose a novel approach integrating GhostNet with an upgraded version of ResNet for osteosarcoma classification. Additionally, we employ an augmented faster region convolutional neural network (FRCNN) for feature extraction, enhancing the system's ability to discern subtle histopathological patterns indicative of osteosarcoma. Furthermore, we introduce the sooty tern optimization algorithm (STOA) to fine-tune the parameters of the deep learning model, thereby boosting classification accuracy. The practical implications of our study extend to telemedicine, mobile healthcare systems, and medical professionals' support tools, where our proposed system offers significant benefits. By leveraging advanced deep learning techniques, our system enables remote diagnosis and consultation, facilitates real-time analysis of histopathological images on mobile devices, and serves as a valuable decision-support tool for medical professionals involved in osteosarcoma diagnosis and treatment planning. In our evaluation, we find that our proposed system outperforms existing models, achieving an impressive accuracy rate of 98%. This performance enhancement, compared to results reported in four previous publications, underscores the efficacy of our approach in accurately identifying osteosarcoma from histopathological images. Overall, our study demonstrates the potential of advanced deep learning techniques coupled with optimized parameterization to overcome challenges in osteosarcoma classification and offers practical solutions with wide-ranging applications in clinical practice and healthcare delivery.