LVM-med-based thermographic image analysis for canine bone cancer detection

Early detection of cancer in veterinary applications is critical for timely intervention and improved treatment outcomes. This paper proposes a novel framework leveraging Large Vision Models (LVMs) for automated analysis of thermographic images to detect bone cancer in canines. Three models, a standard ResNet-50, and two LVM-Med pretrained models (one based on ResNet, the other on a Vision Transformer), were fine-tuned and evaluated on augmented and combined thermographic datasets. Data augmentation significantly improved model generalization. The LVM-Med ResNet variant achieved the highest test accuracy of 93.55% on the combined dataset, outperforming both the ViT-based LVM and the standard ResNet. These findings demonstrate the effectiveness of transfer learning and data augmentation for veterinary thermographic cancer detection and highlight the potential of LVMs to advance non-invasive diagnostic methods in veterinary medicine.