Healthcare is one of the most prospective segments for AI implementation. Artificial intelligence may save doctor’s strengths and time by performing some routine and tiring tasks. Even now, neural networks can detect obesity stages, scan for cancer, and identify a heart attack risk by an image in a more precise and quick way compared to doctors.

Follow the article to get to know about the most successful AI innovations in the healthcare sphere introduced in 2018-2019.

So, how can a doctor benefit from the neural network?

Alzheimer's identification by image

A California-based research team headed by Jae Ho Sohn has created artificial intelligence that identifies Alzheimer's faster than doctors.

When dementia progresses, the level of metabolic activity of brain cells drops, which can be seen on tomographic images. However, the same symptoms are frequent for patients with some cognitive impairments, which complicates the diagnostics process. Artificial intelligence may be helpful in this case.

The scientists were training a neural network feeding it with brain images of 1002 patients – 90% of them were used for training of artificial intelligence while other 10% went for testing. Brain images of other 40 patients made within 2006-2016 were also used for training. The neural networks diagnosed a disease, and the results were compared with doctor’s decisions.

As a result, artificial intelligence was correct at disease diagnosing in 100% of cases while doctors – only 57%. However, compared to the experts, AI showed worse results examining healthy people. The neural network made the right diagnosis in 82% of cases while doctors came up with 91% of correct answers.

Depression recognition by speech

Engineers from the Massachusetts Institute of Technology (MIT) have created artificial intelligence that listens to a person and can diagnose depression based on the obtained results.

Head of the research team Tuka Alhanai claims that a neural network may identify any human emotions – sadness, joy, anxiety as well as depression symptoms. AI is capable of analyzing both spoken and written language.

The neural network was trained on the recordings of the conversations reflecting various emotional states. As shown by the experiments, the neural network managed to recognize depression correctly in 77% of cases. The use of artificial intelligence proved to be more effective than all conventional methods of disease diagnostics.

As explained by the project authors, traditional examination held by psychiatrists requires dealing with a lot of people, which makes it possible to detect the slightest changes in the speech.

Heart attack risk detection

Nils Strodthoff at the Fraunhofer Heinrich Hertz Institute (Berlin) and Claas Strodthoff at the University Medical Center Schleswig-Holstein (Kiel) have created a neural network that is capable of detecting the heart attack symptoms. As stated by the researchers, this development is the first among all neural networks that uses the same methodology as cardiologists do.

One of the most effective methods of heart attack diagnostics is the use of an electrocardiography machine to measure heart activity. A standard ECG device records electrical signals from 12 cables attached to various patient’s body parts. They show heart’s electrical impulses.

The problem lies in increased time consumption. First of all, a doctor has to get the basic signals, ignore any murmurs, and clearly hear a patient’s heartbeat. Then, they are to detect predefined or automatically recognized time intervals and pressure indices of every beat. It the end, the specialist identifies the peculiarities of a heartbeat and classifies its state. The process of classification gets complicated due to a signal called ST segment elevation. Patients with such a signal must be treated as quickly as possible. Those who are not diagnosed with it need to do labor-intensive tests.

Nils and Claas have made this procedure easier. They were training a neural network to scan for a heart attack based on 148 ECG recordings of patients suffering from this disease and 52 healthy people. As a result, the algorithm managed to identify the disease with the same precision as human doctors do but faster.

“The suggested architecture outperforms conventional approaches in this sphere and proves to be as effective as cardiologists,” comment Nils and Claas.

Malignant tumors scanning

Australia-based scientists appeared to be the most precise in this sphere. In 2018, artificial intelligence for the first time ever conducted cancer analysis more successfully than people. The test involved dermatoscopy that implies the examination of skin lesions by their appearance.

Practical dermatologists from 17 countries and a neural network were involved in the experiment. Their task was to identify a melanoma by skin images. The key symptoms of this lesion are dots asymmetry, irregular edges, and a change in color.

The doctors and artificial intelligence worked with 300 photos from the specially designed database. As a result, the neural network correctly defined skin malignant symptoms in 95% of cases while the experienced group of 53 dermatologists – only 88.9%. It should be noted that the algorithm again showed quite worse results at working with healthy patients – both people and artificial intelligence successfully recognized non-dangerous formations (86.6%).

Death prognostics

All the abovementioned neural networks would be helpful at the early stages of the disease. What if a person is death-sick, and to save his/her life, it’s important to recognize any changes in the health state? Due to objective reasons, doctors cannot supervise only one patient. The reason lies in the common problem of medical staff shortage across the globe. American-based manufacturer of the devices in the health segment ExcelMedical in currently working on this issue.

The company has developed neural network Wave Clinical Platform to track biological parameters on 24/7 basis and inform about the possible lethal case. Wave detects even the slightest changes in the organism’s functions and warns the doctors about a possible death 6 hours before.

The algorithm went through the clinical trials at the University of Pittsburgh Medical Center (US). It observed aged people and managed to prevent six deaths.