Enhancing bone cancer detection through optimized pre trained deep learning models and explainable AI using the osteosarcoma tumor assessment dataset

Diagnosis of bone cancer using histopathology images is essential for effective and timely treatment. However, contemporary diagnostic methods struggle to achieve high accuracy and interpretability while utilizing computational methods. Although existing methodologies in deep learning are promising, each suffers from significant limitations that arise from fundamental challenges in hyperparameter optimization, explainability, and generalizability across disparate datasets. Such disadvantages serve as barriers to clinical use, underscoring the need for a more reliable and comprehensible diagnostic framework. In this study, an Optimized Deep Learning Framework for Bone Cancer Detection (ODLF-BCD) algorithm is proposed by jointly combining Enhanced Bayesian Optimization (EBO), deep transfer learning from state-of-the-art pre-trained models (i.e., EfficientNet-B4, ResNet50, DenseNet121, InceptionV3, and VGG16), and explainable artificial intelligence, namely Grad-CAM and SHAP. It mitigates the state-of-the-art limitations through hyperparameter tuning, increased transparency, and data augmentation to balance the dataset. Extensive experiments verify the effectiveness of the proposed framework, where EfficientNet-B4 achieves 97.9% and 97.3% for binary and multi-class classification, respectively. Its performance is also confirmed with high precision, recall, and F1 score. Explainability facilitates the clinical interpretability of model predictions. Then, the proposed framework offers a robust and efficient alternative solution to the C-RAD, automating bone cancer diagnosis and enhancing the accuracy and transparency of the diagnosis. Its potential usefulness could provide clinicians with strong decision support systems for early and precise cancer detection.