A common challenge faced by researchers associated with healthcare institutions is that data of interest are often contained in electronic medical informatics systems that are centered on optimizing clinician/clinician and patient/clinician communication. While this focus naturally enhances the primary goal of care delivery, it is often suboptimal for secondary research purposes. For example at our institution while it is easy for a clinician to view images associated with a specific patient visit, it remains a challenge for an investigator to assemble a cohort of specific images in order to further research objectives. In order to address this important optimization gap we have developed a system for automated image categorization based on a deep neural network. This image classifier organizes the contents of an electronic health record system in a manner which is more amenable to further research by specifically dividing all available images into a lexicon of subclasses. While the current study is focused on dermatology-related images collected by a combined primary and tertiary care center, we expect similar approaches to aid a variety of institutions and clinical specialties.
[1]
Andrew H. Beck,et al.
Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer
,
2017,
JAMA.
[2]
Hongfang Liu,et al.
Ascertainment of asthma prognosis using natural language processing from electronic medical records.
,
2018,
The Journal of allergy and clinical immunology.
[3]
Li Fei-Fei,et al.
ImageNet: A large-scale hierarchical image database
,
2009,
CVPR.
[4]
Sebastian Thrun,et al.
Dermatologist-level classification of skin cancer with deep neural networks
,
2017,
Nature.
[5]
Andrew Y. Ng,et al.
CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning
,
2017,
ArXiv.