Radiologists are once again on the front of new technologies, which they have to deal with. Artificial intelligence may be game changer in the radiological profession, and is coming into clinical practice in the next few years. The purpose of the session is to understand the underlying principles of artificial intelligence, come these tools can help the radiologist in clinical practice and what impact (pros and cons) will have on the profession and the role of the radiologist. The very near use of artificial intelligence in radiology seems inevitable, due to the challenge imposed by quantitative radiology, radiomic, and the huge amount of information (big data) that the radiologist will have to process in the interpretation of the images. Is the time of visual interpretation going down and moving to quantification? Will the radiologist be a data processor, an imaging data manager? The session, with experts in this field, will try to provide a strategic vision on this hot topic.
1. Describe the relationship between artificial relationship, machine learning and deep learning.
2. List the challenges associated with the adoption of a general AI programme that could interpret studies in a manner similar to a radiologist.
3. Explain the reason for the recent excitement about “AI” in diagnostic imaging when the technology has been around for many years.
4. List applications that are not related to image pixel interpretation that can benefit imaging departments that use machine learning.
A variety of difficult challenges in statistics can be solved best with the creation of a “machine” which can provide a simulation or model to discern patterns in a dataset and make predictions. The rapid adoption of “deep learning” in diagnostic imaging has resulted in the development of algorithms for detection, diagnosis, and quantification of medical images at an ever accelerating pace. Major successes in the application of “deep learning/AI” in speech recognition, self-driving cars, translation and strategic games such as Chess, Go, and Poker have resulted in major financial investments and bold and controversial predictions by “experts” about the rapidity of general adoption in radiology and other medical imaging specialties such as pathology, dermatology, and ophthalmology. However, tremendous challenges exist in the implementation of machine learning. The current state of the art would require thousands of algorithms and many millions of imaging studies to replace more than a tiny fraction of the tasks of a diagnostic radiology and would require hundreds of thousands or millions of hours of “expert” time to tag these. Although algorithm development times have been drastically reduced, the time and effort required for testing, verification, and validation of these algorithms in clinical practice has not decreased. Regulatory bottlenecks and medico-legal issues and constraints will need change substantially for widespread adoption within the next several years or decades. Finally, Deep Learning may actually have its greatest initial success in solving non-image related challenges such as image quality, workflow efficiency, improved communication and patient safety.
1. To become familiar with the principles and history of image computing (automated image analysis, artificial intelligence (AI), artificial neural networks (ANN), deep learning).
2. To learn about the opportunities of image computing in radiology.
3. To have a realistic view of the current and future role and impact of image computing to radiology.
The 1970s was the decade that “computed imaging” radically changed the field of radiology. Today “image computing” has become sufficiently mature to have a similar influence on this discipline. In this talk the principles and history of image computing will be shortly summarized and its potential will be illustrated on clinical examples of computer-assisted detection, screening, quantitative measurements, evidence-based diagnosis, early outcome prediction of therapy, and imaging biomarkers. Exploiting this new technology is a logical evolution in the context of value-based health care and should therefore not be postponed or neglected. Instead, it should be considered as an opportunity and embraced by radiologists as an indispensable tool for quality assurance.
1. To become familiar with the basic principles behind deep learning.
2. To understand the requirements and limitations of deep learning.
3. To learn about successful applications of deep learning in radiology.
With deep learning techniques computers can now solve some image interpretation tasks as good as - or even better than - human experts. This lecture explains the basic principles of deep learning and its application in radiology, discusses technical requirements, and presents examples of successful application of deep learning techniques in radiology.
1. To learn how machine learning can help to analyse medical images.
2. To appreciate the need for close cross-discipline collaboration.
3. To understand what are the key ingredients for successful use of today's machine learning technology.
Due to an ever-increasing complexity, large volume of data and high economic pressure the interpretation of medical images pushes human abilities to the limit. Machine learning has emerged as a key technology to address some of the major challenges in medical imaging. In particular, machine learning-based algorithms can be employed to accurately extract quantitative measures such as volumes of anatomical and pathological structures from 3D medical scans, a task that is very difficult to achieve by a human expert. Additionally, ML has potential to discover new patterns of disease, identify abnormalities and extract clinically useful information from high-dimensional, multi-modal imaging data which otherwise would remain hidden.
This presentation will provide an overview of the recent successes and challenges of today's machine learning approaches when applied to the automatic analysis of brain, cardiac and whole-body imaging. Particular focus is put on the interaction between computational and clinical disciplines and we argue for the need of an even closer collaboration between these disciplines and the importance for a common language and understanding of the demands of successful employment of machine learning in clinical practice.