What we can see from the eye alone
Dr. David Feinberg recalls a Google intern’s discovery when learning about machine learning from retina scans.
Video courtesy of HLTH
AI-enabled diagnostics previously thought impossible
In partnership with healthcare organizations globally, we’re researching robust new AI-enabled tools focused on diagnostics to assist clinicians. Drawing from diverse datasets, high-quality labels, and state-of-the-art deep learning techniques, we are making models that we hope will eventually support medical specialists in diagnosing disease. We’re excited to further develop this research towards new frontiers—and to demonstrate that AI has the ability to enable novel, transformative diagnostics.
Using AI to help doctors address eye disease
-
How AI could predict sight-threatening eye conditions
Age-related macular degeneration is the third largest cause of blindness across the globe. If caught early, however, vision loss can be slowed or saved. Our research, published in Nature Medicine, shows how our AI system can recommend—as accurately as world-leading experts—how patients should be referred for treatment for more than 50 eye diseases. Additional research also published in Nature Medicine shows a further improved AI model that has the potential to predict the development of wet AMD within six months.
-
Helping doctors prevent blindness
Our Automated Retinal Disease Assessment, in use in clinics in India and Thailand, shows how an AI can help doctors quickly spot diabetic retinopathy, a leading cause of blindness. With widespread adoption, perhaps millions of patients with diabetes could keep their vision in part to Automated Retinal Disease Assessment assisting doctors. This research was published in JAMA and Ophthalmology. Additional research, published in Lancet Digital Health, showed that we can predict whether patients will develop diabetic retinopathy in the future, which can help doctors customize both treatment and eye screening frequencies for their patients. Learn more
Novel biomarkers for
non-eye related conditions
-
Detecting hidden signs of anemia from the eye
The human eye can reveal signs of underlying diseases like anemia, a condition that affects 1.6 billion people worldwide causing tiredness, weakness, dizziness and drowsiness. In research published in Nature Biomedical Engineering, we were able to use deep learning to quantify hemoglobin levels and detect anemia with de-identified photographs of the back of the eye. This result means it’s possible that someday providers may be able to detect the disease with a simple non-invasive screening tool. Read the post
-
Using computer vision to assess cardiovascular risk
Assessing the risk of cardiovascular diseases is the first and most critical step toward reducing the likelihood that a patient suffers a cardiovascular event in the future. By applying deep learning techniques to retinal images, we’ve been able to reveal factors associated with the risk of a major cardiovascular event like a heart attack or stroke, as published in Nature Biomedical Engineering. This research could help scientists generate more targeted hypotheses and drive a wide range of future research. Read the post
Using AI to improve
lung cancer detection
-
A promising step forward for predicting lung cancer
Lung cancer leads to over 1.8 million deaths per year world wide, accounting for almost one in five cancer deaths, and is the largest cause of cancer mortality. Our research, published in Nature Medicine, shows that deep learning may eventually help physicians more accurately screen for lung cancer and identify the disease even in incidental lung cancer detection workflows. Read the post
Using AI to improve
breast cancer detection
-
Applying deep learning to metastatic breast cancer detection
In our pathology research published in the Archives of Pathology & Laboratory Medicine as well as The American Journal of Surgical Pathology, we showed how a proof-of-concept assistance tool (LYNA) could use deep learning to increase the accuracy of metastatic breast cancer detection. Read the post
-
Using AI to improve breast cancer screening
Despite the wide usage of digital mammography, spotting and diagnosing breast cancer early remains a challenge, leading to delays in detection and treatment. In our research in Nature, we showed how our AI model identified breast cancer in de-identified screening mammograms with greater accuracy, fewer false positives, and fewer false negatives than experts in both the US and UK. Building on this work, we are now conducting a clinical research study to explore whether artificial intelligence (AI) models can help reduce the time from screening mammography to diagnosis, narrowing the assessment gap and improving the patient experience. Read the post
Further research in
imaging and diagnostics
We continue to advance AI-enabled imaging research in other domains, expanding this technology to facilitate transformative diagnostics.
-
Exploring AI advancements in radiotherapy planning to improve efficiency
We are collaborating with Mayo Clinic to study the use of AI to help clinicians plan radiotherapy treatment for cancer. We’ve joined forces to research, train and validate an algorithm to assist physicians with segmenting healthy tissue and organs from tumors to reduce treatment planning time and improve the efficiency of radiotherapy, hopefully allowing clinicians to spend less time planning and more time with their patients. Read the post
-
Using deep learning to help identify skin, hair & nail diseases
We developed a deep learning system (DLS) to address the most common skin, hair and nail conditions seen in outpatient clinical settings, highlighting the potential to assist general practitioners to accurately identify dermatologic conditions. This landmark study was featured on the cover of Nature Medicine. Read the post
-
Using machine learning to detect deficient coverage in colonoscopy screenings
Colorectal cancer (CRC) is a global health problem and the second deadliest cancer in the United States, resulting in an estimated 900K deaths per year. By alerting physicians to missed regions of the colon wall, our algorithm has the potential to lead to the discovery of more adenomas, thereby increasing the adenoma detection rate and decreasing the rate of interval colorectal cancer, as published in IEEE Transactions on Medical Imaging. Read the post
-
Using AI to identify the aggressiveness of prostate cancer
To diagnose the severity of prostate cancer, biopsies are analyzed and given a Gleason grade, which is scored on comparisons to healthy cells. In work published in JAMA Oncology and JAMA Network Open, we explored whether an AI system could accurately Gleason grade prostate biopsies, and our results indicated that the deep learning system has the potential to support expert-level diagnoses. Read the post
