Comparison of machine deep learning with non-medically trained human performance indicated that the machine almost always exceeded acceptable specificity and could operate with higher sensitivity.
The study shows that deep machine learning can be utilized to more accurately identify erythema migrans rashes in early Lyme disease. Recognition of the EM rash is crucial to early diagnosis and treatment. Improved rash recognition using deep learning methodology to prescreen patient rash photos may help prevent later serious manifestations of Lyme disease.
Why was this study done?
There are over 300,000 new cases of Lyme disease annually in the US. In early Lyme disease, blood antibody diagnostic tests are currently unreliable, and the erythema migrans (EM) rash, when present, has varied presentations which can be challenging to accurately identify. Delayed diagnosis and treatment can lead to serious ongoing illness including neurological, cardiac and rheumatologic complications. Accurate recognition of the early presenting EM rash by both patients and clinicians is crucial to facilitating early diagnosis and prompt initiation of appropriate treatment.
The goal of this research is to determine if a machine-based, deep learning approach to analyzing photo images of skin lesions can accurately identify erythema migrans skin rashes of early Lyme disease against other confounding skin conditions. Timely automated prescreening detection of the EM rash could enable earlier diagnoses and treatment and help improve patient outcomes.
How was the study done?
The study developed deep machine learning approaches using artificial neural networks (e.g., DCNNs- deep convolutional neural networks) to analyze visual skin rash photo images to detect and distinguish acute Lyme disease erythema migrans lesions from other confounding skin conditions. The study used 1,834 expert clinician-curated online photo images from unknown individuals with erythema migrans, tinea corporis, herpes zoster, and normal skin. It also included 116 EM lesion images taken of 63 clinically validated research participants from the Mid-Atlantic region. Machine performance was computed against the clinician criterion standard and was tested against a validated set of images and against a panel of non-medically trained individuals.
What were the major findings?
This study found deep learning methodology can classify Lyme disease rashes from patient photos with an accuracy of 86.53%. Results suggested substantial agreement between machine and clinician criterion standard. Comparison of machine deep learning with non-medically trained human performance indicated that the machine almost always exceeded acceptable specificity and could operate with higher sensitivity.
What is the impact of the work?
The study showed our deep machine learning approach can more accurately identify erythema migrans rashes in early Lyme disease. Deep learning is likely a more sensitive prescreen tool than patient self-assessment and has the potential to be more accurate than diagnosis by a general non-specialist physician, who would ordinarily serve as the screening gatekeeper for acute onset EM rashes. Given the frequent under-diagnosis of EM, the use of automated prescreening detection would be beneficial by increasing the number of patients who seek further medical assessment for EM rashes and by minimizing the number of cases with rashes that go unevaluated and undiagnosed. This could facilitate earlier and more accurate diagnoses, earlier and more effective treatments, and help prevent the otherwise serious long-term complications associated with later-stage Lyme disease.
This work was supported by the Johns Hopkins University Applied Physics Lab internal research and development funds as well as the Johns Hopkins University School of Medicine philanthropy funding.
Automated detection of erythema migrans and other confounding skin lesions via deep learning Computers in Biology and Medicine Volume 105, February 2019, Pages 151-156 Philippe M. Burlina, Neil J. Joshia, Elise Ng, Seth D. Billings, Alison W. Rebman, John N. Aucott.
Learn More - https://www.hopkinsrheumatology.org/