Regulatory aspects of artificial intelligence and machine learning-enabled software as medical devices (SaMD)

Automatic correction of performance drift under acquisition shift in medical image classification

“…However, integrated in the workflow it can allow for automatic correction to happen before the next auditing round, as data acquisition updates may happen at any time. Modern AI software as a medical device regulations include the option of 'change protocols' to allow AI devices to improve over time with respect to clinically relevant metrics but must be negotiated and agreed upon with the regulators and include appropriate human oversight in deployment 35 . Thorough validation of processes influencing the performance of a medical device, as well as comprehensive post-market surveillance and continuous model monitoring are critical to the success of this approach in a regulated environment.…”

Evolving Horizons in Radiation Therapy Auto-Contouring: Distilling Insights, Embracing Data-Centric Frameworks, and Moving Beyond Geometric Quantification

“…Naturally, the first and most obvious data-centric strategy involves collecting large data sets that appropriately represent relevant imaging, demographic, and disease-related factors, thereby maximizing model generalizability—a focus strongly emphasized in current regulatory practices. 26 Outside of direct data collection, fields of data-centric AI, like active learning, where models iteratively learn through user interaction, could be used to improve performance and minimize contouring time. Notably, interactive contouring has already been shown to be clinically feasible for HNC tumors 27 and organs-at-risk.…”

“…Furthermore, as we increasingly rely on these models, ensuring that they remain unbiased, particularly toward underrepresented or marginalized communities, is paramount; these issues are becoming increasingly important from a regulatory perspective. 26 The consequences of biased AI can range from inaccurate predictions to reinforcing systemic inequalities. 37 Thus, adopting specific data-centric strategies focused on assuring representation and consistent performance will not just be beneficial—but a moral imperative.…”

Lung Cancer Forecasting Using Hybrid Optimization Technique

“…The use of ML and AI techniques is rapidly expanding in the medical industry. Given the widespread use of AI/ML in risk prediction for many health diseases, as stated in [6,7], it is crucial to evaluate and support practical development of software tools based on AI/ML for early prediction and diagnosis of sickness. Illnesses such as diabetes hypertension [10], COVID-19 [11], hypercholesterolemia [12], COPD [13], systemic lupus erythematosus [17], chronic kidney disease.…”