Abstract
BACKGROUND: Artificial intelligence (AI) and machine learning (ML) are in the process of being integrated into modern healthcare, especially in imaging data-dependent fields like Radiology. Unfortunately, data shows that the acquaintance of physicians and medical students with AI is far from sufficient. Therefore, we implemented an educational chapter within the ASNR MICCAI BraTS 2023 Brain Metastases Challenge. Its purpose is to educate medical students on the basics of AI and ML application in the diagnostic imaging and apply deliberate learning approach to teach biology and physics of MRI imaging of brain metastases. MATERIAL AND METHODS: Pilot group of 15 medical students from EU was expanded to 150, to include radiology residents and research associates. Fundamentals of MR imaging and process of lesion segmentation was taught through lectures and written guides. Trainees were involved in refining preprocessed cases to create ground truth segmentations for AI algorithm training within BraTS-METS Challenge. The annotators received individual support through question and answer sessions and expert reviews. Lectures specifically targeted to AI, ML, ethics, and Open Science introduced trainees to field experts. Specific skills related to building and publishing databases suitable for ML algorithm training, including all necessary preprocessing steps such as quality control, skull stripping, and normalization, were developed throughout the training process. Alternative methods for education, including #BraTSBrainMetsTeachingPearls Twitter series were also implemented which offered in-depth guidance on interpreting the imaging features of brain metastases and the ability to interact with imaging content. RESULTS: Participants reported an enhanced understanding of neuroradiology, particularly in identifying imaging features of brain metastases. Their gained proficiency in ML/AI empowers them to efficiently assess existing models or replicate the workflow of model training. The early clinical exposure and extensive ML/AI lectures were highly valued, providing deeper insights into the applications and limitations of these technologies. The educational strategy combining constructivist principles with practical experience proved highly beneficial, equipping them with skills relevant to both research and clinical practice. CONCLUSION: Insufficient integration of AI and ML in medical education necessitates effective and early training. Adopting a constructivist approach has equipped students with vital skills in Open Science, database development, and model training, preparing them to fully utilize these technologies while understanding their limitations.