The computer is taught to predict 5-year survival rate of patients

Taking computer tomography (CT) of the bodies of patients and treating them with the help of special software, the scientists were able to predict 5-year survival rate of patients with an accuracy of almost 70%. This was reported in a new study published in Scientific Reports. Lead author Dr. Luke Ogden-Rainer (Luke Oakden-Rayner) from the School of public health (School of Public Health) at the University of Adelaide (The University of Adelaide) and colleagues believe that the new data can be useful in the development of personalised medicine.

National institutes of health USA (National Institutes of Health) define personalized medicine as “a new approach to the prevention and treatment of diseases, taking into account differences in genes, environment and lifestyle of the people.”

As the authors of a new study individualized medicine based on the discovery of biomarkers that are accurate predictors of the risk of disease development, treatment response and prognosis. Scientists believe that an important role in the new approach can play radiology:

“…we believe that the images obtained during conventional radiological examinations, long ignored in the context of personalised medicine. We believe that the use of a powerful modern machine-learning techniques in application to radiological images could lead to the discovery of new informative biomarkers.

Recent developments in the field of imagery analysis demonstrated that the results of the automated image processing for many diseases at the accuracy comparable with the results of the biopsy, microscopy, and even DNA analysis”.

In the study, Dr. Oakden-Rayner and his colleagues decided to find out whether it is possible to train the computer to recognize the computer tomography thus, to predict 5-year survival rate of patients. For starters, scientists have collected more than 15 thousand images of seven different tissues, including heart and lung. All MRIs belonged to the patients aged 60 years and older. Using the method of logistic regression, the researchers identified a number of traits associated with a 5-year survival rate. The researchers then combined the data obtained with the technology of deep learning, suggesting that the computer will “learn” to recognize images.

“Computers are able to combine large amounts of data and highlight subtle details,” explains Dr. Oakden-Rayner.
In the next phase, the scientists used “trained” the computer to perform chest CT 48 patients aged 60 years and older. The researchers found that the automated system is able to predict 5-year survival rate with an accuracy of 69% compared to the forecast made by medical specialists.

“Although the study used a small sample, our work shows that the computer has learned to recognize symptoms of disease on the images. In order to teach doctors, need special intensive training,” adds Dr. Oakden-Rayner.

The group plan of the research is to test the developed techniques on tens of thousands of patient images. But the authors now say that their work demonstrated the possibility of using machine learning and computed tomography in the development of personalised medicine. In particular, the new method can be used for early detection of serious diseases that require specific treatment.