Closing Remarks

If you were able to get here after completing all the tasks above - congratulations! You have gone through the challenging process of integrating knowledge of clinical context, data analysis, machine learning systems, and medical imaging networking to create a fully functional AI module for a radiological system.

Armed with this knowledge you will be able to get quickly started with a vast majority of problems in 3D radiological imaging space, and even transfer this knowledge over to non-radiological modalities that generate 3D images.

At the moment of writing in 2020, medical imaging AI is a very rapidly growing space, and the potential of the field is staggering. We are only starting to get access to good clinical datasets, the ImageNets of medical imaging is yet to come, clinician researchers are just starting to wrap their heads around what is possible with machine-learning-based technology and tools are becoming better every day. Information flow between data scientists and clinicians is key to unlocking the potential of medical AI and helping clinicians reduce the amount of mundane work, become more precise, efficient, and less stressed. This is just the beginning.

Further Resources

If you are curious to learn more about the space and see what others are doing, here are a few useful resources, companies and societies to watch for.

Conferences and professional societies

MICCAI Society hosts an annual conference dedicated to medical imaging and related fields, and also hosts a number of challenges. One that has consistently generated good volumetric datasets is called BRATS
Radiological Society of North America is a renowned organization that unifies medical imaging professionals around the globe. In addition to hosting the eponymous largest medical imaging conference in the world it has been turning more attention to AI recently, and hosted interesting medical imaging competitions within its "AI challenge" program. Last year's challenged featured a classification problem for CT imaging (although with the focus on 2D methods)
SIIM is a society that focuses on medical imaging informatics and it has recently started running a machine learning sub-conference called C-MIMI

Academia

It wouldn't be much of an overstatement to say that almost every academic medical center in the world is running some sort of a medical imaging AI program. These are all very interesting since they are rooted in clinical expertise and benefit from access to data. They vary in size and often are a part of larger, disease-specific programs. A couple efforts worthy of noting are:

Center for Clinical Data Science by Parthers Healthcare
Stanford's AIMI
National Consortium of Intelligent Medical Imaging, kicked off by the University of Oxford and the UK's National Health Service

Startups

There are plenty and there will be more. Some choose to pursue a clinical workflow, some focus on application of particular machine learning technique and some capitalize on existing clinical footprint and invest in platforms that accelerate others' efforts. Some established players are:

Cortechslabs - focuses on quantitative analysis of brain images. Of particular note is the software called Neuroquant which uses deep learning to produce reports with MRI-based volumetric measurements of structures inside brain that are related to age-related neurodegenerative disorders such as Alzheimer's. Sounds familiar? :)
Mirada Medical - Oxford-based company that advanced a field of radiation oncology with its deep-learning-based segmentation models
Arterys - Silicon Valley startup that was the first to obtain an FDA clearance for a deep learning medical imaging suite for oncology.
Enlitic - San Francisco-based company aiming at diagnostic use cases that accelerate radiologic workflow
Nuance is a Boston-based company that produces a well established platform of choice for radiological dictation. Recently the company focused a lot of effort on a marketplace for medical imaging AI solutions where startups that do not quite have Nuance's reach can deploy their software.
Terarecon - similarly to Nuance, this Californian company started in core diagnostic radiology and expanded with an AI marketplace offering branded "EnvoyAI"

Big Tech

Some big cloud providers are eyeing the space closely, and running their own programs and projects related to medical imaging.

Microsoft Research has a project InnerEye that for the past 10+ years has been exploring the use of machine learning for a variety of medical imaging applications. One of the instructors of this course had the honor of spending a significant part of his career as a team member here.
Google DeepMind is a group within Google doing some cutting-edge AI research, including some work on medical imaging. We can credit them with the contribution to the invention of the U-net which has been prominently featured in this course.