Masayuki Tsuneki1. 1. Medmain Research, Medmain Inc., Fukuoka, Japan; Division of Anatomy and Cell Biology of the Hard Tissue, Department of Tissue Regeneration and Reconstruction, Niigata University Graduate School of Medical and Dental Sciences, Niigata, Japan. Electronic address: tsuneki@medmain.com.
Abstract
BACKGROUND: Deep learning is a state-of-the-art technology that has rapidly become the method of choice for medical image analysis. Its fast and robust object detection, segmentation, tracking, and classification of pathophysiological anatomical structures can support medical practitioners during routine clinical workflow. Thus, deep learning-based applications for diseases diagnosis will empower physicians and allow fast decision-making in clinical practice. HIGHLIGHT: Deep learning can be more robust with various features for differentiating classes, provided the training set is large and diverse for analysis. However, sufficient medical images for training sets are not always available from medical institutions, which is one of the major limitations of deep learning in medical image analysis. This review article presents some solutions for this issue and discusses efforts needed to develop robust deep learning-based computer-aided diagnosis applications for better clinical workflow in endoscopy, radiology, pathology, and dentistry. CONCLUSION: The introduction of deep learning-based applications will enhance the traditional role of medical practitioners in ensuring accurate diagnoses and treatment in terms of precision, reproducibility, and scalability.
BACKGROUND: Deep learning is a state-of-the-art technology that has rapidly become the method of choice for medical image analysis. Its fast and robust object detection, segmentation, tracking, and classification of pathophysiological anatomical structures can support medical practitioners during routine clinical workflow. Thus, deep learning-based applications for diseases diagnosis will empower physicians and allow fast decision-making in clinical practice. HIGHLIGHT: Deep learning can be more robust with various features for differentiating classes, provided the training set is large and diverse for analysis. However, sufficient medical images for training sets are not always available from medical institutions, which is one of the major limitations of deep learning in medical image analysis. This review article presents some solutions for this issue and discusses efforts needed to develop robust deep learning-based computer-aided diagnosis applications for better clinical workflow in endoscopy, radiology, pathology, and dentistry. CONCLUSION: The introduction of deep learning-based applications will enhance the traditional role of medical practitioners in ensuring accurate diagnoses and treatment in terms of precision, reproducibility, and scalability.