A Modern Approach to Osteosarcoma Tumor Identification Through Integration of FP-Growth, Transfer Learning and Stacking Model

Abstract

The early detection of cancer through radiographs is crucial for identifying indicative signs of its presence or status. However, the analysis of histological images of osteosarcoma faces significant challenges due to discrepancies among pathologists, intra-class variations, inter-class similarities, complex contexts, and data noise. In this article, we present a novel deep learning method that helps address these issues. The architecture of our model consists of the following phases: 1) Dataset construction: advanced image processing techniques such as dimensionality reduction, identification of frequent patterns through unsupervised learning (FP-Growth), and data augmentation are applied in this phase. 2) Stacking model: we apply a stacking model that combines the strengths of two models: convolutional neural networks (CNN) with transfer learning, allowing us to leverage pre-trained knowledge from related datasets, and a Random Forest (RF) model to enhance the classification and diagnosis of osteosarcoma images. The models were trained on a dataset of publicly available images from The Cancer Imaging Archive (TCIA) [12]. The accuracy of our models is evaluated using classification metrics such as Accuracy, F1 Score, Precision, and Recall. This work provides a solid foundation for ongoing innovation in histology and the potential to apply and adapt this approach to broader clinical challenges in the future.