ML And AI In Medicine: Areas, Technologies, And Perspectives

The idea of using artificial intelligence in medicine dates back to 1972 when Stanford University’s MYCIN became operational. It was a prototype AI program used to study the issue of blood contamination. Early AI research continued mostly at U.S. institutions (MIT-Tufts worked together, and Stanford and Rutgers University actively developed the technology). In the 1980s, Stanford University continued its work in AI with the Stanford University Medical Experimental Computer for Artificial Intelligence in Medicine (SUMEX-AIM).

Thanks to the growth in computing power and the emergence of new artificial intelligence technologies, work in this area has become much more active. Regularly, there is news about the next scientific discovery made with the help of neural networks and machine learning. What interesting things can be said about the possibilities and prospects of AI in medicine today?

AI In Radiology

A lot of medical imaging data is abundantly stored in small, localized systems. But what if you utilize deep learning by uploading the data to the cloud and “feeding” it to AI? Machines and algorithms can effectively interpret the imaging data, identifying patterns and anomalies.

The most obvious use case is a radiologist/assistant engaged in identifying and locating suspicious skin masses, lesions, tumors, internal hemorrhages, masses on the brain, etc. The computer works faster and more accurately and, therefore, is able to give specific data about the disease after a few seconds of information processing. A human being cannot do this.

There is another point. Highly qualified specialists are expensive, and their demand is huge. They are under tremendous pressure, literally getting bogged down in the data streams that come at them from all sides. According to this research, such a specialist should give a diagnosis every 3–4 seconds. Artificial intelligence can enhance an ordinary specialist’s skills, helping him understand complex situations. Thereby reducing false diagnoses and saving lives.

Detection of rare or difficult-to-diagnose diseases often depends on the experience of the doctor, as well as the disease progression. Simply put, it may not be recognized until it’s out in the open. Training a computer on large datasets containing raw images and many forms of pathologies associated with certain diseases makes it possible to improve the quality of diagnoses and the number of diseases detected. This is the idea being developed by startup AIDOC.

AI can improve medical institutions’ work quality by automating a time-consuming and responsible part of doctors’ work. With the help of computer algorithms, it is also possible to monitor the effectiveness of treatment.

Microsoft InnerEye project is a good example of such technology. It should also be noted that MRI and other advanced imaging systems used for early cancer detection work with ML.

With AI, for example, prostate segmentation can be performed, or several different images (e.g., ultrasound, CT, and MRI) can be combined to get a more accurate picture. Artificial intelligence can also recognize cancer during routine medical procedures and even surgery (it often happens that another malignancy goes undetected during surgery).

AI In Pathology

Pathology diagnosis involves examining a slice of tissue under a microscope. Using Deep Learning to train an image recognition algorithm combined with human expertise will provide a more accurate diagnosis. Analyzing digital images at the pixel level can help in detecting pathological changes that the human eye can easily miss. And it will provide more effective diagnostics.

Such technology is being developed, for example, by Harvard Medical School. The algorithm uses speech and image recognition technology to recognize images with pathologies and trains computers to distinguish between cancerous and non-cancerous growths. Combining this algorithm with human performance resulted in 99.5% accuracy.

We also recommend you read our article Artificial Intelligence (AI) In Healthcare: Use Cases and Risks.

Machine Learning And Medical Science

Petabytes of data are generated in all sorts of medical settings. This data, unfortunately, is usually haphazardly scattered and unstructured.  This is by no means a rebuke to physicians. They don’t have to treat as much as they have to report on treatment. But chaos is a great hindrance to planning and global surveillance of the health of a country or the world at large.

An additional challenge is that, unlike standard business data, patient data does not lend itself well to simple statistical modeling and analytics. A powerful AI-enabled cloud platform with access to medical databases can effectively analyze mixed information (e.g., blood pathology, genetic features, X-rays, medical history). It is also (theoretically) capable of analyzing input data and identifying hidden patterns that are not visible due to too much medical information.

Interpretable AI models and distributed machine learning systems are well suited for these tasks. They will advance medical science by finding new patterns and racial/gender/age-specific patterns in people and generate more accurate data on the health status of populations in specific regions.

We recommend you read our article 5-Step Guide to Healthcare Mobile Application Development.

Surgical Robotic Assistants

Already, many surgeries are being performed using computer vision and manipulators controlled by the surgeon. This is a significant part of the development of medical technology, leveling the factor of human fatigue and increasing the efficiency of procedures.

AI robots have the potential to help conventional surgeons in a big way. For example:

• Monitor the doctor’s work, acting as an insurance policy in case of inattention.
• Improve visibility for the surgeon, reminding him or her of the sequence of actions during the procedure.
• Create precise, minimally invasive tissue incisions.
• Reducing pain for the patient by selecting the optimal incision geometry and suture.

But for a successful realization of such a project it is necessary to accumulate experience. Develop software for robot-surgeon interaction. Gather an array of information based on real surgeries (both human-only and human+robot).

A good option may be the generation of virtual reality space by the computer to control the surgeon’s actions in real-time. It is also possible to use telemedicine and remote surgery to perform relatively simple operations.

We recommend you read our article Top 10 Healthcare Trends & Innovations in 2024.

Virtual Nurses

Virtual nurse technology is also being actively developed, capable of “accompanying” the patient throughout the entire period of treatment. AI is planned to be used to monitor the patient’s condition, record the indicators of sensors installed on the patient’s body, and provide answers to the patient’s standard questions (about the time of procedures, doctor’s name, duration of treatment, etc.).

An additional function of such assistants can be a transcription of the doctor’s voice messages into text. You know how much time medics spend on all sorts of writing. Voice recording in patients’ medical records would increase the time spent directly on treatment by getting rid of excessive paperwork.

It should be emphasized that this is not a modernized chatbot. This is a nurse assistant, which establishes a quality communication channel between a medical institution, a doctor, and a patient.

Perhaps, such AI will be placed outside of hospital rooms. For example, if a person is treated at home. This reduces the number of unnecessary contacts and eliminates the need for regular visits to the hospital.

Another plus of AI is that computers can take over diagnostic functions in regions where there is a shortage of doctors. People in any situation will be provided with the minimum necessary medical care. And this is important.

Creating New Medicines With AI/ML

The creation of new medicines with the help of AI has been mentioned many times. This is a really promising idea. Pharmaceutical giants are increasingly using AI technologies to solve the hellishly difficult problem of successfully creating new effective drugs against cancer, metabolic disorders, and immune problems. You can read about how they’re doing it, for example, here.

Computer algorithms, using the experience and information already accumulated by humans, are able to simplify the process of finding cures. They are used to:

• Screening out ineffective and highlighting promising drug compounds at an early stage of development.
• Creating and identifying new drug combinations that have been overlooked/discarded as ineffective earlier.
• Identifying mechanisms of action of a particular drug in simulated situations.

Biopharmaceutical company NuMedii has already created AIDD technology that uses BigData and AI to rapidly detect links between drugs and diseases at a systemic level. Many medical startups are also testing AI capabilities to analyze multi-directional and unstructured medical data (research papers, patents, clinical trials, patient histories).

Precision Medicine And Disease Prevention

Precision medicine term refers to a fundamentally new approach to disease treatment and prevention that considers a person’s genetic and lifestyle characteristics, as well as the environment.

Simply put, treatment will not be based on average standards that fit most people who get sick. People are too different, and the same pill works differently for everyone. An individualized therapy plan will be drawn up, allowing for a faster, safer, and more effective cure.

The task is exciting but difficult. We must find how to treat a person based on their medical history, lifestyle, genetic data, and constantly changing test results.

Artificial intelligence can calculate why and under what circumstances diseases are more likely to occur. And prepare doctors to intervene (on a case-by-case basis) before a person even shows the first symptoms. Cardiovascular and oncological problems, diabetes – if we teach AI to anticipate these diseases, humanity will become healthier, and life expectancy and quality should also improve.

A less precise variant is also possible. For example, when AI detects the degree of similarity of the clinical picture in patients with the same diseases. This data makes it possible to categorize patients separately and determine which treatments best suit them. AI automatically assigns them to a specific group and makes treatment recommendations by analyzing the information about a patient when they see a doctor.

Predicting Pandemics

Medical organizations are using ML technologies to monitor and predict potential epidemic outbreaks that could span different parts of the world. By collecting data from satellites, monitoring real-time social media posts, and analyzing other important information from the internet, these tools can accurately predict an outbreak in a particular region. This can be a boon for third-world countries lacking a developed healthcare system. However, for developed countries, the function of predicting the increase in disease incidence will not be excessive.

An example is the medical application ProMED-mail. With its help, the WHO can monitor the situation in the world and forecast disease outbreaks in real-time.

The technology is constantly being improved. Scientists want to make the monitoring tool more efficient and self-sufficient. For example, they want to predict a malaria outbreak based on temperature and average monthly precipitation.

Conclusion

AI will move deeper into medicine and this is inevitable. The early you incorporate AI in your processes or healthcare organization, the bigger advantage you will have over your competitors.

Search for a reliable IT partner for your healthcare project? Contact us at OS-System!