Tag Archives: deep learning

Deep learning in Radiology and Pathology affects Bronchoscopists

Photo by Andrew Neel, on Unsplash

This is a second post relating to the promising role of artificial intelligence in interventional pulmonology.  My point is that lung specialists will spend less time learning facts and figures that are easily replaced by computer-generated analyses of complex algorithms. Much of this is because of Deep learning

This subset of machine learning (programs that adjust themselves as they are exposed to more data, but without human input) uses artificial neural networks (algorithms built on unstructured data). The word deep is a technical term referring to the number of layers in the neural network. Artificial Neural networks being a set of algorithms modeled after the human brain and used to recognize patterns.  Image recognition is one example, and its principles are responsible for much of the work done today in radiology and pathology. 

For example, using deep learning and pattern recognition, AI reveals CT abnormalities and interprets findings (Google’s AI team recently outperformed traditional radiologists looking at 45,800 screening CTs for lung cancer https://www.fiercebiotech.com/medtech/google-s-cancer-spotting-ai-outperforms-radiologists-reading-lung-ct-scans), and chest radiographs are accurately interpreted using fuzzy logic interpretations of spatial relationships (https://www.ijcaonline.org/specialissues/dia/number1/4156-spe320t).

Pathology is another area where practice patterns will undoubtedly change. In many regions, expert cytologic interpretation of lung and mediastinal nodal specimens is lacking. Digital pathology (image-based information generated from a digital slide) allows real-time interpretation by computers at sites that are distant from wherever the procedure takes place. Humans already do this despite the cost and logistic difficulties.  I believe that artificial intelligence will soon facilitate and universalize the process (https://www.healthimaging.com/topics/artificial-intelligence/ai-lung-cancer-slides-accuracy-pathologists). 

In today’s post, my goal was to introduce the concept of deep learning and provide a few examples of how this mode of artificial intelligence will affect procedural practice by changing how chest radiology and pathology are practiced. Rather than devote study time to learning X-ray and cytology interpretation, future bronchoscopists will improve their abilities to incorporate findings into appropriate management plans, as well as communicate results to patients, caregivers, and health-care teams.

  • Please subscribe to Colt’s Corner to automatically receive email notification of future posts. Sign up with your name and email on the NEWSLETTER button on the Bronchology International home page at  www.bronchoscopy.org.