Algorithm for predicting mortality
Computer trained to predict 5-year survival of patients
Anna Stavina, XX2 century, based on Medical News Today: Patient mortality could be predicted through computer analysis of organs
By taking computed tomograms (CT) of patients' organs and processing them with the help of special software, scientists were able to predict the 5-year survival rate of patients with an accuracy of almost 70%. This is reported in a new study published in the publication Scientific Reports (Precision Radiology: Predicting longevity using feature engineering and deep learning methods in a radiomics framework). The lead author of the work, Dr. Luke Oakden-Rayner from The School of Public Health at The University of Adelaide and his colleagues believe that the new data may be useful in terms of the development of individualized medicine.
The US National Institutes of Health defines individualized medicine as "a new approach to the prevention and treatment of diseases that takes into account differences in genes, environment and lifestyle of specific people."
According to the authors of the new study, individualized medicine is based on the discovery of biomarkers that are accurate predictors of the risk of developing the disease, response to treatment and prognosis. Scientists believe that radiology can play an important role in the new approach:
"... we believe that the images obtained during a routine radiological examination have been ignored for a long time in the context of individualized medicine. We believe that the use of powerful modern machine learning technologies in the application to radiological images can lead to the discovery of new informative biomarkers. Recent developments in the field of image analysis have demonstrated that the results of automated image processing in many diseases are comparable in accuracy to the results of biopsy, microscopy and even DNA analysis."
In the course of the study, Dr. Oakden-Rayner and his colleagues decided to find out whether it was possible to train a computer to recognize CT scans in such a way as to predict the 5-year survival rate of patients. To begin with, scientists have collected more than 15 thousand images of seven different tissues, including cardiac and pulmonary. All tomograms belonged to patients aged 60 years and older. Using the logistic regression method, the researchers identified a number of signs associated with 5-year survival. The scientists then combined the data obtained with deep learning technology, suggesting that the computer itself will "learn" to recognize the images.
"Computers are able to combine large amounts of data and highlight subtle details," explains Dr. Oakden–Rayner.
At the next stage of the work, the scientists used a "trained" computer to analyze CT scans of the chest of 48 patients aged 60 years and older. The researchers found that the automated system is able to predict 5-year survival with an accuracy of 69% in comparison with the forecast made by specialist doctors.
Out of 25 patients assigned by the program to the low-risk group,
2 people died in 5 years, 20 out of 23 in the high–risk group.
Graph from an article in Scientific Reports – VM.
"Although a small sample was used for the study, our work demonstrates that the computer has learned to recognize the manifestations of diseases in the images. In order to teach this to doctors, special intensive training is needed," adds Dr. Oakden–Rayner.
The research group plans to test the developed technique on tens of thousands of patient images. But the authors of the study already claim that their work has demonstrated the fundamental possibility of using machine learning and computed tomography in the development of individualized medicine. In particular, the new technique can be used for early detection of serious diseases requiring specific treatment.
Portal "Eternal youth" http://vechnayamolodost.ru
06.06.2017