ESPEYB21 2. Antenatal and Neonatal Endocrinology Artificial Intelligence (1 abstracts)
2.7. Transforming neonatal care with artificial intelligence: challenges, ethical consideration, and opportunities
Sullivan BA , Beam K , Vesoulis ZA , Aziz KB , Husain AN , Knake LA , Moreira AG , Hooven TA , Weiss EM , Carr NR , El-Ferzli GT , Patel RM , Simek KA , Hernandez AJ , Barry JS & McAdams RM
J Perinatol 2024 Jan;44(1):1-11. doi: 10.1038/s41372-023-01848-5
Brief Summary: This paper discusses the potential use of artificial intelligence (AI) in neonatology for clinical practice and research. The authors highlight the importance of multi-stakeholder involvement, and the need for well designed protocols to not only test outcomes but also to address the ethical issues involved and usability, bias, transparency and acceptability.
We have been using AI since the 1950s when automated ECG interpretations became available using computer algorithms. Machine learning (ML) has taken AI further, using algorithms that iteratively learn from patterns in large datasets. It can be done either using labeled datasets (supervised ML) or unsupervised, which may uncover unexpected data patterns. At time of writing, a PubMed search on use of AI in healthcare reveals 20,080 publications, half of these appearing since 2022.
According to the authors, neonatology can benefit from AI due to the large wealth of clinical data generated for each infant, many of whom have complex conditions and multiple comorbidities. By linking data from clinical examination and fetal monitors with laboratory and imaging results (as well as genetic and proteomic data), prediction models can be tested for significant comorbidities, such as necrotizing enterocolitis and sepsis. AI may also help to address care of bronchopulmonary dysplasia in extreme premature and SGA babies, such as finding novel risk factors, assessing treatment strategies and predicting long-term disease burden.
Image analysis is another potential domain for AI. Retinopathy of prematurity typically requires detailed manual retinal examinations. Interpretation of retinal images by AI could decrease inter-observer variability and improve diagnostic accuracy and efficiency. Similarly, MRI imaging is often performed at term to assess brain injury in neonates with previous abnormal transcranial ultrasounds or significant comorbidities, but it has low predictive value. AI studies are examining use of computer-generated algorithms to predict both short and long term clinical outcomes.
There are multiple challenges to implementation of AI in neonatology. These include data acquisition (unbalanced or incomplete data), data processing (anomaly detection and cleaning), and data analysis and testing. All steps require quality assurance, including accuracy and reproducibility, and also ethical use of data, transparency and avoidance of bias. Users must understand the decision-making models. Government oversight will be a sine qua non, but regulatory bodies will be faced with constantly evolving and complex software and hardware. Given the costs of AI implementation and healthcare in general, perhaps the biggest challenge will be avoiding inequality of access to AI and to the wealth of information that can be mined and used to improve healthcare.
Volume 21
Article tools
My recent searches
My recently viewed abstracts
Authors
ESPE Yearbook of Paediatric Endocrinology
ISSN 1662-4009 (Online)
© Bioscientifica 2024 |
Privacy policy |
Cookie settings
BiosciAbstracts
Bioscientifica Abstracts is the gateway to a series of products that provide a permanent, citable record of abstracts for biomedical and life science conferences.
Find out more