How deep learning may revolutionize image-based diagnosis

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon

For years, medical researchers and vendors have tried their hand at involving computer aided diagnosis (CAD) into patient care. However, slow integration has allowed for newer cognitive technologies such as deep neural networks, or deep learning technology, to find an onramp into radiology imaging.  

Cognitive technologies expert Tom Davenport delves into deep learning technologies and artificial intelligence (AI) in a recent article in Forbes.   

Specifically, Davenport asserts that new cognitive technologies, such as deep learning, have the potential to improve medical imaging in pathology labs.  

The Center for Clinical Data Science (CCDS) at Partners Healthcare in Boston, Davenport explains, is centered around bringing deep learning technologies from research labs into clinical practices. Massachusetts General Hospital and Brigham & Women's Hospital in Bosten are currently in a joint effort with CCDS with a goal to improve clinical practice within the two hospitals and the general healthcare system  

"Its [CCDS] goal is to employ machine learning and other artificial intelligence technologies to improve the healthcare delivery system; in particular, a key CCDS objective is to improve the effectiveness of imaging-based diagnosis," Davenport said.  

According to Davenport, deep learning is a better technology than CAD. Deep learning technology has the ability to identify the most important images for radiologist to examine, rather than clinicians being swamped with images to choose from with a CAD approach.  

Passing along verbatim from CCDS Executive Director Mark Michalski, MD, any amount of success deep learning technology may have in patient care requires:  

  • Industry partnerships 
  • Labeled data  
  • Integration with clinical care flows 
  • Leading-edge data science skills 

"The objective of harnessing the power of deep learning for medical image analysis, and embedding it in an effective program of clinical care, is one of the most important challenges in artificial intelligence," Davenport said. "To achieve that objective will require collaborative and long-term work by groups like researchers at Partners, many other hospitals and physicians, the ACR, deep learning startups, providers of labeled images, vendors of imaging technology, and the FDA."  

The CCDS is trying to develop deep learning prototype products or "use cases" to ease the technology in clinical space, according to Davenport.  

Read the full article here.