Articles of Interest: AI-Enabled CTA Radiation Reduction, the Future of Drug-Eluting Stents, and Explainable Auto-Encoders

< Back To Feed

Deep Learning Image Reconstruction reduces radiation dose by 43% in CCTA scans.

A recent study by Benz et al published in European Radiology compared a deep learning image reconstruction tool (DLIR) with a conventional Adaptive Statistical Iterative Reconstruction-Veo (ASiR-V) and produced some surprising results.  

Abstract image of an artificial neural network creating a heart

The study involved 50 patients undergoing 2 CCTA’s, one with a normal dose, and one with a 40% reduction in dose. The scans were reconstructed with ASiR-V and DLIR respectively. The study concludes that DLIR enabled a radiation dose reduction of 43% without significant impact on image noise, stenosis severity, plaque composition, and quantitative plaque volume. By reducing radiation exposure by more than 40% while preserving all relevant clinical information by the simple application of a new deep learning image reconstruction algorithm, the authors provide a great demonstration of the potential of ML algorithms to improve patient outcomes.

Q&A: Past, present, and future of drug-eluting stents.

A recent Q&A with Dr. Amar Krishnaswamy from the Cleveland Clinic caught our attention. In the interview, Dr. Krishnaswamy, section head of invasive & interventional cardiology at the Cleveland Clinic, recalls the history of drug-eluting stents and where future trends may take us.  

He recalls that, even in the last decade, stents have made several major breakthroughs. With more companies producing and researching the technology, he expects to see more innovation in bioabsorbable scaffolds, in-stent restenosis, and intravascular imaging. 

New findings for interpretation of EHR autoencoders.

A study by Chushig-Muzo et al published in the journal Artificial Intelligence in Medicine demonstrated a new approach to increasing interpretability of AE (auto encoder) based models. Interpretation of deep learning-based algorithms is one of the major barriers that stand in the way of clinical AI adoption and effective support of the clinical decision-making process. Since AE models are based on nonlinear transformations they result in black-box models, which inherently reduces interpretability. However, Chushig-Muzo et al used real-world data from electronic health records to show how their approach can identify patterns in the data to improve interpretation. We are excited to see how approaches like this can improve the adoption of AI models in clinical decision-making.

< Back To Feed