Swin Transformer for Bone Metastasis Classification: A Comparative Study with Transfer Learning Models

Abstract

Bone metastasis is a severe complication in lung cancer patients, significantly impacting prognosis and treatment planning. Accurate and timely diagnosis is crucial for improving patient outcomes. In recent years, deep learning has demonstrated remarkable potential in medical image analysis, particularly in disease classification. This study presents a comparative analysis of deep learning models for bone metastasis classification, evaluating four convolutional neural network (CNN) architectures-EfficientNetB0, MobileNetV2, ResNet50, and VGG16-alongside Swin Transformer, a vision transformer-based model. The models were trained and tested on a dataset comprising 767 training images, 164 validation images, and 166 test images, categorized into benign, malignant, and normal cases. Performance was assessed using accuracy, precision, recall, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score metrics. Experimental results indicate that Swin Transformer outperformed all CNN-based models, achieving the highest classification accuracy of 95 %, demonstrating superior performance across all evaluation metrics. Among CNNs, MobileNetV2 and VGG16 achieved the highest accuracy (94 %), with MobileNetV2 being computationally more efficient. ResNet50 and EfficientNetB0 exhibited lower accuracy, particularly in distinguishing benign cases. These findings highlight the potential of Swin Transformer for enhanced bone metastasis detection and suggest the need for further research on model generalization using larger and more diverse datasets.