A Data Scientist’s Experience in Decoding Chest Imaging by Vidya MS

The Chest Imaging Update 2018 held by the Narayana Health group, brought together over 150 radiologists, pulmonologists and doctors gathered to update and improve their knowledge in the reporting of Chest Imaging, both X-ray and CT. As a data scientist with keen interest in medical imaging, my aim was to get an inside look into the daily practice of medical professionals in detection and diagnosis of pulmonary diseases.

The conference opened with Dr. Vimal Raj, Conference chair and an accomplished Cardiothoracic Radiologist, stating that thoracic imaging is not reported well enough and there is a lot more value that a radiologist could add in the diagnosis to enable better treatment options for the patient. The flow of the subsequent sessions was extremely well constructed reflecting the everyday reporting workflow in the chest imaging space – starting from assessment of the humble chest X-ray to the CT followed by diagnosis and treatment plans. The second day was mainly focused on hands on sessions.
The first session of the day was conducted by Dr. Kishore Kumar from NH and Dr. Aparna Irodi from CMC Vellore on what never to miss on a chest X-ray and assessment of neonatal chest-X-rays. Dr. Kishore covered the important parts to access in a chest X-ray before reporting it as a normal one. One of the challenges that he brought out was that these areas are often easily missed and there is large amount of misdiagnosis made while assessing a chest X-ray which could led to significant impact later on for the patient. One case was presented where an early stage opacity, which could have been easily caught, progressed into a cancerous stage. There are a multitude of reasons for misdiagnosis, including, superimposition of thoracic structures, similarity in radiographic appearances of some chest diseases, and subtlety of some chest pathologies rendering them indistinguishable. Each chest X-ray takes a trained radiologist several minutes to review and given the ubiquitous nature of the X-ray in the imaging world, this often leads to significant increase in the workload. Another reason could be the variance of reporting across radiologists due to inconsistent terminologies.
The sessions then shifted towards the more complex CT imaging, where the physics and protocols behind the CT imaging were first introduced, followed by multiple sessions on reporting of lung cancer on a CT and standard definitions of chest patterns on CXR and CT to ensure better reporting.
Since the conference was focused on updating radiologists and how to better report their studies, the important question on why accurate reporting is needed was excellently handled by Dr. Murali Mohan. He walked through a comprehensive series of statistics conducted by various institutions, to explain why the importance of reporting accurately.
One of the highlights and most interesting sessions for the day for me, was the ‘Multi-Disciplinary Team’ panel featuring, Dr. Vimal Raj, Dr. Murali, Dr. Aparna, Dr. Rajani Bhat, Dr. Ranganatha R. They wonderfully presented how an MDT team conclude in diagnosis and determining a course of action when things are unclear. This was extremely insightful for an outsider like myself to truly understand the complications, uncertainties and how the various disciplines come together to define the best course of action for the patient.
Considering all the complications mentioned above, the variations in manifestations of the same medical condition across demographics, overlap of multiple patterns and mimic cases, many of these cases require significant expertise in the diagnosis of a thoracic scan. There is also a high variance in reporting across radiologists and the opinions are sometimes subjective especially in chest X-rays. Many of these uncertainties sometimes require a group of expertise to reach a consensus. In addition to these clinical challenges, developing an AI system to aid the reporting process requires large amounts of consistently reported data to truly learn the patterns on an image. Many of these systems also do not have access to much of the other patient information that may be pivotal in assessment. The difficulty in obtaining such large high-quality data makes it more complex to build these systems.
AI is still not there to diagnose the myriad of pathologies on a thoracic scan, primarily the chest X-ray though there have been many solutions claiming to detect a limited set of abnormalities. That being said, with the immense improvement in technology in the recent years, AI can certainly aid a radiologist in diagnosis and subsequently the treatment. The focus for AI models should be around aiding radiologists in repetitive tasks, ensuring that highly skilled expertise focuses on more complex and abnormal cases, detecting and highlighting unusual patterns in the image (especially those than can be easily missed), and also in providing clinical insights to a radiologist in the detection of abnormal regions.

Finally, I would like to thank the organizers of the Chest Imaging Update 2018. Overall, it was a wonderful experience for me to be in the midst of a large group of experts from various disciplines, interact with some of them to understand the key problem areas. 

The article was first published on the Author’s LinkedIn pulse Blog, its been re-published here with the Author’s permission. All ideas presented are author’s personal views

Vidya S.M

Data Scientist at Philips. Data Scientist with a demonstrated history of working in the hospital & health care industry. Skilled in AI, image processing, & algorithms. Strong engineering professional with a Master’s degree focused in Machine Learning & AI from National University of Singapore.

This site uses Akismet to reduce spam. Learn how your comment data is processed.