At the Radiological Society of North America’s (RSNA) 2016 annual meeting at McCormick Place Convention Center in Chicago, Illinois, on November 27 through December 2, Lunit will showcase Data-driven Imaging Biomarker (DIB), a novel AI-powered visual perception technology derived from cutting-edge deep learning technology applied with large-scale medical data.
In contrast to conventional Computer-Aided Detection (CAD) that is heavily dependent on guidance (lesion annotation) of radiologists, DIB lets the machine define important diagnostic features by itself solely from large-scale data. This can be possible by employing feature-learning capability of deep convolutional neural network.
At the event, Lunit will exhibit the first public demonstration of DIB for chest radiography and mammography at booth #4074 in South – Hall A.
DIB for chest radiography serves to act as a “second reader” in detection and differentiation of abnormal lesions in chest x-rays, including pneumonia, emphysema, diffuse interstitial lung disease, tuberculosis, lung cancer, and pulmonary metastasis, with high diagnostic performance. DIB is expected to help both general practitioners and radiologists interpret CRs more efficiently and accurately.
DIB for Mammography is designed to improve breast cancer detection rates in mammography, mainly focused on breast cancer detection in dense breast, an aspect of breast cancer screening widely regarded to be most challenging. Although it is in the early stages of development, Lunit’s DIB for mammography aims to 1) detect and localize lesions, 2) suggest their BI-RADS categories, and 3) predict their malignancy probabilities.
DIB will also be covered in the following presentations:
“Data-driven Imaging Biomarker: Uncovering Diagnostic Features from Large-scale Medical Images Using Deep Learning” –IN023-EC0-SUB: Informatics Sunday Poster Discussions, Sunday, November 27, 1:00-1:30 PM
“Performance Assessment of Data-driven Imaging Biomarker for Screening Pulmonary Tuberculosis on Chest Radiographs” --IN210-SD-MOB: Informatics Monday Poster Discussions, Monday, November 28, 12:45-1:15 PM
“Applying Data-driven Imaging Biomarker in Mammography for Breast Cancer Screening” --SSE02-06: Breast Imaging (Quantitative Imaging and CAD), Monday, November 28, 3:00-4:00 PM
Deep Convolutional Neural Network Approaches in Making a Diagnosis with Chest Radiographs: Initial Experience --SSJ05-06: Chest (Functional), Tuesday, November 29 Tuesday 3:00-4:00 PM
Lunit is looking forward to meet worldwide healthcare providers who are interested in DIB solutions.
To schedule a demonstration of DIB and/or meeting at the upcoming RSNA 2016, please reach out to us via email@example.com
Nov. 22, 2016, 7:33 a.m.
On Oct. 17, Tumor Proliferation Assessment Challenge was held as a satellite event of MICCAI 2016 and Lunit won first place in all of the three tasks in the challenge. Unlike the previous challenges, participants were asked to predict the tumor proliferation score that can be integrated into current prognostic grading systems, being more relevant to actual clinical practice.
Lunit’s researchers at the challenge venue: Sunggyun Park, Kyunghyun Paeng and Sangheum Hwang (from left)
Led by Kyunghyun Paeng, co-founder and research scientist at Lunit, the team designed a system based on the latest advances in artificial intelligence to help pathologists understand the histological slides more accurately and efficiently.
Top 5 challenge results are shown (click for more information). Semi-automatic methods are omitted as they are out of the challenge scope. The entries with asterisks(*) are teams that used separate additional data for training., not provided in the challenge.
Not only is tumor proliferation, highlighted by number of dividing cells, or mitotic cell count, important in the diagnostic process of histological slides especially in breast cancer, but it is also an important biomarker that is highly predictive of prognosis, or how long a patient will survive. This has very important clinical implications, as the type of treatment selected may be dependent on such biomarkers.
Accurate and consistent assessment of tumor proliferation on whole slide imaging level will bring much value to digital pathology. Lunit's successful endeavors mark a step forward toward application of state-of-the-art software analytics in digital pathology to bring unprecedented clinical benefits to patients.
Detailed code and description of methodologies used for this challenge will be released in the near future.
Oct. 27, 2016, 1:33 p.m.