AI in Brain Tumour Detection: Comparative Analysis of YOLOv10 and PaliGemma2 with Public Perception Insights in Bosnia and Herzegovina

Abstract

Artificial Intelligence (AI) has emerged as a significant advancement in healthcare, particularly in medical imaging, where it assists in detecting abnormalities, segmenting lesions, and predicting conditions. However, comparative analyses between conventional object detection models and emerging vision-language architectures remain limited, especially for brain tumour detection. This study addresses this gap by comparing two algorithms: YOLOv10, a widely used object detection model, and PaliGemma2, a newer vision-language model that integrates image and text modalities. Despite the rapid development of AI tools, comparative studies evaluating their effectiveness and public acceptance in healthcare remain scarce. Therefore, this study aims to evaluate the technical performance of AI algorithms in brain tumour detection and to assess societal readiness for their adoption in Bosnia and Herzegovina. Performance was assessed using a labelled magnetic resonance imaging (MRI) dataset and evaluated through accuracy and precision metrics, while public perception was analysed through a survey involving 344 participants. The results indicate that YOLOv10 consistently outperformed PaliGemma2, likely due to its optimisation for object detection tasks, whereas PaliGemma2’s multimodal design required greater computational resources. The findings from the survey revealed positive public acceptance of AI in healthcare, accompanied by calls for greater education, careful implementation, and appropriate professional training. Overall, results from this study provide empirical evidence supporting the practical applications of AI models in medical imaging and highlight the importance of integrating ethical and educational frameworks for AI adoption in developing healthcare systems, such as in Bosnia and Herzegovina.