Deep learning, a type of machine learning that involves learning representations of data, has been an area of interest for healthcare and life science companies from the study of human genomes to helping doctors gather information from patients about their medical conditions to spotting errors in medical images. Machine learning is also used in computer assisted detection and diagnosis products for medical imaging, such as screening for breast cancer. One day machine learning technologies could be widely used for clinical decision support for medical images.
A new study from Google researchers published in the Journal of the American Medical Association found that a set of deep learning algorithms were able to detect with a high level of sensitivity and specificity diabetic retinopathy and macular edema in a series of images of the back of the eye.
For clarity, sensitivity refers to the ability to detect a disease and specificity refers to the ability to recognize when a disease is not present.
To train the deep learning algorithms, the study used 128,000 images, which were each evaluated by three to seven U.S. board-certified ophthalmologists.
Two sets of images, one containing 9,963 images from 4,997 patients and a second containing 1,748 images from 874 patients were used for the study. The algorithms scored 97.5 percent and 96.1 percent for sensitivity and scored 93.4 percent and 93.9 percent for specificity for detecting referable diabetic retinopathy, respectively. The study defined referable diabetic retinopathy as moderate or worse diabetic retinopathy or referable macular edema, according to a panel of at least seven U.S. board-certified ophthalmologists.
The finding is significant because for patient populations with a “substantial disease,” high sensitivity and high specificity are essential to ensure that false positives and false-negative results are minimal. The results also were comparable to manual detection of diabetic retinopathy by ophthalmologists.
A Deep-dive Into Specialty Pharma
A specialty drug is a class of prescription medications used to treat complex, chronic or rare medical conditions. Although this classification was originally intended to define the treatment of rare, also termed “orphan” diseases, affecting fewer than 200,000 people in the US, more recently, specialty drugs have emerged as the cornerstone of treatment for chronic and complex diseases such as cancer, autoimmune conditions, diabetes, hepatitis C, and HIV/AIDS.
The study noted that some of the big picture benefits of automated grading of diabetic retinopathy include improving efficiency for ophthalmologists, reducing barriers to access and enabling early detection and treatment, thereby improving outcomes. Although the study noted the usefulness of machine learning for spotting patterns in data, most of the healthcare applications have focused on computing characteristics specified by experts. That’s led to algorithms to detect specific lesions or to predict the presence of any level of diabetic retinopathy, according to the study.
A little more than 28 percent of Americans have diabetic retinopathy, the study noted. There are 415 million diabetic patients worldwide at risk for developing the condition which can lead to irreversible blindness if it is not caught at an early stage when the disease can be treated, a Google Research blog highlighting the research study noted.
The study concluded that more research is necessary to determine the feasibility of applying this algorithm in a clinical setting and to determine whether the use of the algorithm could lead to improved care and outcomes compared with current ophthalmologist assessment methods.
Photo: Bigstock
