Deep learning methods applied to routinely collected chest CT scans can predict 5-year mortality rate in individuals aged over 60.
This study showed that the use of deep learning techniques (“convolutional neural networks”) to assess tissue changes in routinely collected medical images can extract surrogate biomarkers for overall individual health and latent disease.
"Instead of focusing on diagnosing diseases, the automated systems can predict medical outcomes in a way that doctors are not trained to do, by incorporating large volumes of data and detecting subtle patterns," Dr Oakden-Rayner says.
There was no significant difference between the accuracy of the deep learning model and a human-defined feature model. However, as a proof-of-concept experiment, it is promising to see that the classification results using the deep learning approach is capable of predicting 5-year mortality with comparable prediction accuracy to the current ‘human engineered’ approach.
Given the widespread use of medical imaging, with further refinement and testing on larger datasets, the use of deep learning models to predict medical outcomes can be readily translated to clinical use with little additional cost. Dr Oakden-Rayner also notes that "Our research opens new avenues for the application of artificial intelligence technology in medical image analysis, and could offer new hope for the early detection of serious illness, requiring specific medical interventions."