Developing reliable AI tools for healthcare

In a new technical blog post, TecH enthusiasts, Diakun Research, have presented their latest research on the potential of using predictive AI in healthcare. The researchers used data from UK medical images, specifically mammography datasets, to investigate the accuracy of predictive AI tools for interpreting medical images. They showed that by combining human expertise with predictive AI, CoDoC, a system designed to help interpret these images, could potentially improve the reliability of AI models for real-world medical use. The research was conducted using de-identified and historic clinical data, with the prediction made based on a pre-trained AI model and a clinician’s opinion. CoDoC required only three inputs for each case in the training dataset and could be inserted into a hypothetic future clinical workflow involving both an AI and a clinician. Despite being theoretical, the researchers showed that combining the best of human expertise and predictive AI resulted in greater accuracy than with either alone. CoDoC was able to achieve a 25% reduction in false positive results for a maommography dataset, improving accuracy across various scenarios. Their findings underscore the potential benefits of harnessing the benefits of AI in combination with human strengths and expertise. This research was conducted using data from UK medical imaging equipment, which suggests that CoDoC’s potential to adapt to various settings is promising. The researchers recognize that there is still a need for robust evaluation before adopting AI-driven healthcare solutions in real-world settings. However, with rigorous evaluations of these tools and systems, we can help ensure they are safe and beneficial to users.