47 Medical Computer Vision Computer Vision News in two public health facilities in Kenya, that highlighted some of the perceived attractions, misconceptions and expressed hesitancies of introduction of AI-enabled low-cost ultrasound-based gestational age estimation in this novel setting. She argued that medical computer vision (MCV) for low-and-middle-income countries (LMICs) is more than just a data application area due to unique real-world characteristics such as limited data size, high class imbalance and data heterogeneity. This has required, for instance, design of few-shot learning solutions, domaininvariant segmentation methods, and use of statistical priors to guide anatomy detection. She described the importance of emerging areas as well, presenting early results of using interpretability (ProtoPNet) for gaining clinician and patient trust in algorithm decisions, and of classimbalanced federated learning-based analysis to enable collaboration between international research sites when data cannot be shared due to privacy and legal concerns. From Aasa Feragen’s presentation:
RkJQdWJsaXNoZXIy NTc3NzU=