Date: 30th May 2019
The prevalence of deep learning in the healthcare industry is gaining momentum and this branch of AI is becoming a sophisticated tool to analyse large data sets, in a quick and efficient manner. Often reported to out-compete its human equivalent in both speed and accuracy it has been hailed as the future of diagnostic testing. Deep learning’s predictive powers can also be applied to a wide range of clinical settings, from assessing the risk of bleeding after surgery to predicting childhood asthma and remission, even the risk of mortality can be modelled.
Whilst it is becoming increasingly difficult to counter gainst deep learning as a potent analytical tool it does have its limitations, however, and Hongfang Liu and colleagues from the Mayo clinic having been assessing these confines in a paper published in Digital Medicine. Using previous studies approved by the Mayo clinic they have challenged deep learning models against older more ‘traditional’ machine learning models to predict patient outcomes. The results; that newer is not always better. What the study does highlight is that each model has strengths and weaknesses and that if any of these AI tools are to integrate fully and successfully into the healthcare industry it will be crucial to compare and assess the contribution of this growing range of methods to suitably match technique to application.
Chen, D., S. Liu, P. Kingsbury, S. Sohn, C. B. Storlie, E. B. Habermann, J. M. Naessens, D. W. Larson and H. Liu (2019). “Deep learning and alternative learning strategies for retrospective real-world clinical data.” npj Digital Medicine 2(1): 43.
https://doi.org/10.1038/s41746-019-0122-0
