Feb 22, 2021 — 3 min read
Our demo website (https://insight.lunit.io/) for Lunit INSIGHT CXR and Lunit INSIGHT MMG has over 2,000 users around the world. With the pandemic, more and more users are trying our AI solutions online and contacting us for further product information.
Recently, we received a delightful email from one of our users. Prof. M. Erkin Aribal, M.D., from the Acibadem MAA University in Turkey, conducted his research on the potential clinical value of Lunit INSIGHT MMG using our demo website. And his study has been accepted as a recorded presentation to the European Congress of Radiology 2021.
For a sneak preview of his presentation, check out his email interview below. He kindly explained to us how he planned his research using our demo website, and what he has found out from his study.
How did you come across our demo website?
I met Lunit during the Radiological Society of North America 2019. During my meeting at Lunit booth, I was introduced to the online demo website. I figured I could develop my idea of planning a study with breast cancers in the screening project that we had organized for five districts in Istanbul county.
Tell us about your experience with our demo website.
The website was easy-to-use and user-friendly. The AI result appeared very fast and highlighted areas of concern in colors, indicating the probability of the existence of the finding.
What did you aim to achieve from your study and how did you design it?
We aimed to evaluate the diagnostic performance of the AI algorithm in detecting breast cancers, particularly missed and interval cancers. We prepared for datasets of screening mammograms consisting of normal mammograms, and mammograms with screen detected, including missed and interval breast cancers. And those images were retrieved from PACS.
What is the key takeaway of your study?
The AI algorithm reached AUC levels between 0.840 to 0.903, a similar accuracy rate to that of radiologists. It successfully detected a number of missed and interval breast cancers. The finding suggests that Lunit AI certainly has the potential to augment the diagnostic performance of breast cancer detection as a second reader.
How is the breast screening environment in Turkey?
The breast cancer rate below age 50 is quite high in Turkey. The national breast screening program starts at age 40. Two readers interpret mammograms and a third reader is assigned for inconsistencies.
We have three drawbacks to mammography screening. First of all, lack of experienced breast specialists is a major limitation. Secondly, the low participation rate of women to the national screening program hinders interval breast cancer detection. The participation rate is as low as 15%, especially in rural areas. A majority of women have their screening mammogram for only once in their lifetime. Thirdly, dense mammograms account for 45% of the entire screening mammograms, which causes low sensitivity in breast cancer detection.
How would AI help Turkish radiologists improve their diagnostic performance of breast cancer detection?
As mentioned above, the lack of experienced breast specialists and the importance of having a low rate of interval cancers requires meticulous and accurate breast cancer detection in the national screening program.
Lunit AI proved to detect a number of missed and interval breast cancers and to be efficient as a second reader in our study. It will help radiologists increase their reading speed and accurately interpret abnormal mammograms.
Do you have any other plans to conduct research using AI?
We are planning to conduct further studies where AI is used in decision-making for other imaging procedures such as ultrasound and tomosynthesis.