Deep Learning (DL) is capable of identifying features in images, enhancing image quality, and spotting outliers and abnormalities. These same abilities are now being transferred to healthcare applications.
While the concept of AI has been around since the mid-1950s, the brainchild of research scientists at Dartmouth meant to address problems that were difficult for humans to solve but simple for computers, it was not until the 2010s that AI really began to flourish and impact our daily lives.
Similarly, one of the earliest Machine Learning applications in medical imaging came to the fore back in 2013 when Google DeepMind‘s team decided to throw their weight behind the medical opportunities of their technologies after winning an astounding series of games against the world’s best living Go players, an abstract strategy board game.
It is only very recently however that Deep Learning algorithms have been introduced that are able to learn from examples and prior knowledge. Although these technologies are not yet ready to be rolled out on a large scale, they are moving ever closer to more accurate and quicker diagnoses via deep learning-based medical imaging.
Radiology is arguably the area of medicine where Deep Learning has had the biggest impact on medical imaging. The technology is already improving patient care drastically, all while saving time and money for healthcare organisations. However, many professionals in the industry are concerned that the technology may one day replace them.
These concerns were addressed in a report by the Association of University Radiologists Radiology Research Alliance Task Force
“Some may perceive deep learning algorithms as a threat to medicine and radiology. However, deep learning is like any other tool, intrinsically neither good nor evil, but rather dependent on the application.”
However, similar concerns have been raised in the past surrounding other AI technologies and have subsequently been debunked.
The AI healthcare revolution is being driven by an ever-increasing abundance of data. A report from the EMC states that health data reached 153 Exabytes in 2013 and is expected to grow 48% annually, reaching 2000+ Exabytes by 2020. The challenge for organisations now lies in extracting value from this data. There is plenty of room for advancements in technology particularly if this data is to be made readily available HIE's and with close to 80% of this data remaining unstructured, there is a lot of work yet to be done.
Deep Learning requires these large sets of data with which to make predictions based on learning. In medical imaging, this translates to early diagnoses of diseases. These advancements have already been achieved in Alzheimers detection and will continue to aid medical professionals in making early diagnoses, drastically improving the survival rate.
Here are three of the most promising applications of deep learning in medical imaging
1. Tumor Detection
A prime example of this can be taken from the detection of Melanoma which is the deadliest form of skin cancer, with 5 million cases each year in the United States.
One of the most promising near-term applications of automated image processing is in detecting melanoma
To detect the tumor, the DL algorithm learns important features related to the disease from a group of medical images and then makes predictions based on that learning. - says John Smith, senior manager for intelligent information systems at IBM Research.
The ability of deep learning to improve diagnoses of melanoma was highlighted in a recent report in which Deep learning outperformed 136 of 157 dermatologists in a head-to-head dermoscopic melanoma image classification task.
In the report, it was stated that a convolutional neural network (CNN) received enhanced training with 12,378 open-source dermoscopic images. In a head-to-head comparison, the CNN outperformed 136 of 157 participating dermatologists. The CNN was capable of outperforming dermatologists of all hierarchical subgroups (from junior to chief physicians) in dermoscopic melanoma image classification.
Medical imaging can also be used for non-invasive monitoring of disease burden and effectiveness of medical intervention, allowing clinical trials to be completed with smaller subject populations and thus reducing drug development costs and time.
A new tool revealed in 2013 employs Deep Learning to reveal changes in tumor images, enabling physicians to determine the course of cancer treatment. “The software can, for example, determine how the volume of a tumor changes over time and supports the detection of new tumors,” said Mark Schenk from Fraunhofer MEVIS
2. Improving MRI's
Magnetic resonance imaging (MRI), also known as nuclear magnetic resonance imaging, is a scanning technique for creating detailed images of the human body. A technology first developed in the 1980's it is firmly established as one of the preferred diagnostic imaging tools.
There are advancements coming to market now, which are driven by Deep Learning.
Arterys, a DL medical imaging technology company, enables a much more efficient visualization and quantification of blood flow inside the heart, alongside a comprehensive diagnosis of cardiovascular diseases, thus making it possible for cardiac assessments on MR systems to occur in a fraction of the time of conventional cardiac MR scans.
3. Reducing human error
The effectiveness of medical imaging relies on both image and interpretation quality. The interpretation of images is mainly the responsibility of the Radiologist, these are prone to error due to factors like fatigue, distractions and even biases. This is one reason patients sometimes have different interpretations from various doctors, which can make choosing a plan of action a stressful and tedious process.
Lunit, a South Korean startup established in 2013, uses its DL algorithms to analyze and interpret X-ray and CT images. Lunit’s system is able to provide interpretations in 5 seconds and with 95 per cent accuracy, an achievement that has attracted investments of $2.3 million through international startup incubation programs in just 3 years.
Another South Korean startup Vuno, is helping doctors and hospitals to combat disease by putting medical data to work. Vuno uses its ML/DL technology to analyze the patient imaging data and compares it to a lexicon of already-processed medical data, letting doctors assess a patient’s condition more quickly and provide better decisions.
The future of Deep Learning in medical imaging.
Deep learning in medical imaging is widely regarded to be a critical component in the fight against cancer. Dr. Bradley Erickson from the Mayo Clinic in Rochester, Minnesota, believes that most diagnostic imaging in the next 15 to 20 years will be done by computers.
As part of this effort in the ‘war on cancer’, Google DeepMind has partnered with UK’s National Health Service (NHS) to help doctors treat head and neck cancers more quickly with DL technologies. The research is being conducted in coordination with the University College London Hospital.