- VIT_Conceptz
In this study, we employed three deep learning models ResNet50, MobileNetV2, and DenseNet-121 to perform concept detection, which involves identifying and locating relevant concepts in medical images. This provides the foundation for generating coherent captions in the subsequent caption prediction. In medical imaging, concept detection plays a pivotal role. It enables accurate disease diagnosis and monitoring by identifying specific features, such as tumors, fractures, or anomalies. These concepts guide treatment planning, ensuring timely interventions. Among the models, ResNet50 achieved the highest performance, followed by MobileNetV2 and DenseNet-121. These results indicate that ResNet50 is the most effective model for identifying relevant concepts within medical images. This study provides insights into the applicability of different convolutional neural networks for medical image analysis, contributing to advancements in automated medical image captioning. Our team secured 8th place on the overall challenge leaderboard in the Concept Detection Task of the 8th edition of Caption Challenge in ImageCLEFmedical 2024