Computer vision (CV) is an umbrella term to describe the algorithms and hardware necessary to replicate human sight and to understand the role of the objects in focus. In healthcare, it has countless applications, from a faster and more accurate diagnosis to surgery assistance or enhancing the life quality of visually impaired patients. The long-term goal of using this technology is to give doctors a way to spend less time analyzing results and allow them to spend more time bringing added value to the patient in terms of advice and care.
For example, computer vision software behind a visual search engine can both find a dress similar to one from the Oscars or be retrained to identify pictures with tumors. A program able to segment images into layers to detect different objects in a room can also be successfully used for splitting a CT scan into layers for mental illness recognition or organ identification.
How Does Computer Vision Work?
Each project where the computer needs to see like a person is unique. For example, a self-driving car has different requirements from software for cancer detection. Some of the principles guiding these endeavors as well as the code powering them are similar, just the training datasets are different.
To a computer, an image is a series of numbers. The role of machine learning is to help computers make sense of the sequences of numbers and identify similarities which lead to classifications. A first resemblance with the human brain is using context to make the right assumptions.
Like human judgment, these models are not perfect. The way a computer recognizes an image is by breaking it down in smaller bits and comparing these bits with an existing library, trying to find matches. It makes assumptions and seeks to minimize the difference between the premises and the actual image by comparing the two and aiming for as little deviation as possible.
Computer Vision Applications in Healthcare
There are a few significant areas where CV can prove useful in healthcare both immediately and soon enough, as the technology becomes more widely adopted. Here’s a small selection of these areas below.
Computer Vision for Medical Imaging and Diagnosis
From detecting cancer by taking a picture with a smartphone to finding abnormalities in internal organs, computer vision can be trained to pay attention to particular traits.
The immediate benefit of using such a system is getting a diagnosis fast and in a totally non-invasive way, as the images necessary for the algorithm can be generated by a CT scan or an MR scan. For example, 4D Flow shows doctors the blood flow and heart functioning in real time. Furthermore, the algorithm was trained with labeled scans, and it can make the difference between a healthy patient and a dysfunctional heart.
Applications in Surgery
Post-partum hemorrhage can be fatal if the mother is not getting the right blood replacement as soon as possible. Until now, the evaluation of the amount of lost blood was just a visual approximation by skilled nurses. Computer vision included in the Triton system takes the guesswork out of the equation and offers a scientific answer in less than 10 seconds.
As the results were validated on new mothers, the system will also be used during C-section surgeries. The outcomes mean that the patients will get just the right amount of blood transfer during their operations, thus decreasing the risks for complications and the total time spent in the hospital.
Helping the Visually Impaired
Computer vision can make guiding dogs and white canes a thing of the past. A lightweight camera could pick up images from the environment, send them to a processor that through computer vision determines the type of objects in the proximity, creates a navigation route and directs the person by vibrations or verbal feedback.
The limitations of such a system are mostly connected to the network bandwidth and the processor, which means that it could have difficulties handling very dynamic environments. For now, these smart glasses can be a suitable alternative for indoor environments.
Treatment Plan Monitoring
The most common reason why treatment plans fail is that patients don’t follow them exactly as described by their physicians. Sometimes they forget to take their dose, take it too late, or take a different amount from the prescribed one.
Computer vision can combine two technologies to help in fighting against these problems. The first one is facial recognition to make sure the user is the right person. Next, through object recognition, the system can scan the drugs the patient takes, and in connection with a scheduling app, it can ensure that it happens at the right time. Such a technology can be embedded in a smartphone app and helps clinical trial efforts and patients who are struggling to keep up with their treatment plans at home.
“Seeing” Mental Conditions
The prospect of a camera filming you and assessing your mental health is terrifying, and at the moment not even feasible or morally acceptable. However, determining neurological illnesses by simply looking at a CT scan is already being tested at Mount Sinai Hospital. The algorithm was trained with over 30K images to identify acute neurological problems in patients. However, such a tool doesn’t come cheap and requires a hefty investment in infrastructure as well.
The Watchful Eye of AI
As technology progresses, it will become common practice to use computer vision for medical diagnostics. We can even expect further democratization and the launch of medical-grade apps available on app stores, much like we now have fitness apps. The challenge right now for medical organizations striving to leverage these technologies is to find those providers who can offer them certified solutions that have been consistent in rendering top results.