Image: Images of the proximal left anterior descending coronary artery most strongly predict mortality (L) and survival (R) cases (Photo courtesy of the University of Adelaide).
A new study suggests that analysis of computerized tomography (CT) images of internal organs could predict 5-year mortality with almost 70% accuracy.
Researchers at the University of Adelaide (UA; Australia) and other institutions conducted proof-of-concept experiments to demonstrate how routinely acquired cross-sectional CT imaging may be used to predict patient longevity as a proxy for overall individual health and disease status, using computer image analysis techniques. To do so, they first gathered 15,957 CT images of seven different tissues from patients aged 60 and older; using logistic regression, they identified a number of image features that were linked to 5-year mortality.
Based on the human-defined image features, they then used machine learning and a range of radiomic classifier models that included convolutional neural network random forests, support vector machines, and boosted tree algorithms in order to teach a computer to make 5-year mortality predictions. They found that as expected, the random forest model performed the best on the human-defined feature classifiers. An analysis showed the results were comparable to clinical methods for longevity prediction. The study was published on May 10, 2017, in Nature Scientific Reports.
“Recent advances in the field of medical image analysis have shown that machine-detectable image features can approximate the descriptive power of biopsy, microscopy, and even DNA analysis for a number of pathologies,” concluded lead author Luke Oakden-Rayner, PhD, of the UA School of Public Health, and colleagues. “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.”
Deep learning, a computer learning method which automatically discovers visual features that are suited to a specific task through a process of optimization, has rapidly overtaken more traditional methods in many computer vision tasks, such as image recognition and segmentation, and have approached or even surpassed human level capabilities for complex “real-world” tasks such as image recognition, speech recognition, natural language processing, complex game playing, and more.
University of Adelaide