AI and Medicine
Data quality and accuracy
Medical AI systems are designed to analyse large amounts of medical data, such as electronic health records, medical images, and genetic data, to identify patterns and make predictions about a patient's health. The accuracy of these predictions is heavily dependent on the quality of the data used to train the AI system. If the data used to train the AI system is incomplete, inconsistent, or incorrect, the system's predictions will likely be flawed. For example, if an AI system is trained on medical data that is not representative of the population it is intended to serve, it may be biased for certain groups of patients. Likewise, if the data available does not cover a broad enough demographic, there is the danger of applying incorrect assumptions to different patient groups. Similarly, if the data is not properly labelled and annotated, the AI system may not be able to distinguish between relevant and irrelevant data, leading to inaccurate predictions. Moreover, ensuring the quality and accuracy of data for medical AI is crucial because these systems are often used to make critical decisions that affect people's health and well-being. For instance, medical AI is used to help diagnose diseases, develop treatment plans, and predict patient outcomes. If the data used to train the AI system is of poor quality, the system's predictions may not be reliable, potentially resulting in incorrect diagnoses, inappropriate treatments, and harmful outcomes for patients. To address these issues, organizations must implement robust data governance policies and procedures that ensure the quality and accuracy of the data used to train and test medical AI systems. This includes ensuring that the data is properly acquired, stored and labelled. Further, the data should be verified for completeness, consistency, and correctness, implementing safeguards to ensure data privacy, security, and compliance with ethical and legal requirements. By doing so, organizations can ensure that the medical AI systems they develop are accurate, reliable, and safe for patients. Oversight of the data included in the first instance will minimize these effects, and help to train the AI quickly and efficiently. Buy-in from healthcare professionals to verify this data in the first instance will ensure this process increases pace effectively as the AI ‘ learns ’ from the data inputted.
Ensuring the reliability of AI
AI, like any other form of medical technology, needs to be reliable. To do this safeguards and best practice guidelines need to be implemented to ensure the best possible outcome for the patients. This consistency can be ascertained in many ways. Robustness testing: Testing the AI models under a wide range of conditions to ensure that they perform regularly and reliably. This also allows researchers to determine the limitations of the models to provide best practice guidelines for clinicians to follow. Fig. 4 is a Roth curve depicting the accuracy of an AI at diagnosing hypoplastic left heart syndrome. The green line is the relevant data and the important take away from this graph is that the closer the green line is to the top left the better the AI is at diagnosing. These results are promising. However, when discussing them at an online seminar on innovations in foetal cardiology, Dr Tom day concluded that ( Figure 4 - Roth's curve depicting accuracy of an AI at diagnosing hypoplastic left heart syndrome- Day,2022)
37
Made with FlippingBook - PDF hosting