[Techtonic 2021] Efficiently and effectively developing medical imaging AI systems - Patrick Bangert

Detecting and diagnosing diseases based on medical images is one of the major applications of artificial intelligence (AI) in the next 3 – 5 years. We report on largely automating the human annotation process of medical images, which is the time-consuming and expensive prerequisite to AI modeling. Through a novel image sorting method, a novel unsupervised structure learning framework, a novel unsupervised object detection method, and a novel pixel-labeling based approach, we have lowered the necessary human labor far below prior state of the art. After labeling, we train sophisticated deep learning models to perform the diagnosis making use of automatic machine learning (AutoML) approaches as well as using multiple computers simultaneously to reduce the run-time of training workflows.

We present these methods in a high-level overview and practical applications to real-life datasets from our industry partners on breast cancer, colon cancer, brain tumors, and intravascular ultrasound.

#AI #DeepLearning #ActiveLearning #Healthcare #MedicalImaging

※ Techtonic은 삼성SDS가 연구개발 중인 최신기술 소개는 물론, Cloud, Security, SW Engineering, Customer Experience 등 다양한 분야의 전문가들이 모여 생생한 경험과 교훈을 공유하는 자리입니다.


----------------------------------------------------------------------------------------

🤗 아래의 더욱 다양한 채널에서
      클라우드 전문 기업, 삼성SDS를 만나보세요! 🤗

🔹 홈페이지: https://www.samsungsds.com/
🔹 네이버 포스트: https://post.naver.com/my.naver?membe...
🔹 블로그: https://samsungsds.com/kr/story/index...
🔹 페이스북: Facebook: samsungsds
🔹 트위터: Twitter: ictstory