Improving Healthcare with
Machine Intelligence

Another disease that is primarily diagnosed visually is skin cancer. Researchers from Stanford University have applied similar AI techniques as in the previous example to train AI to identify skin cancer on images [7]. For this purpose, they have used around 130,000 clinical images and compared the performance of the AI system to diagnoses made by more than 20 dermatologists. The system achieved performance on par with all tested experts demonstrating that AI is capable of classifying skin cancer with a level of competence comparable to expert dermatologists. This insight opens exciting new opportunities to extend the reach of dermatologists outside of the clinic: if smartphones are outfitted with an AI system as described above, every smartphone owner can scan her skin to identify potential skin cancer and, in the case of a positive result, contact her doctor promptly. Since many who currently notice an unusual spot on their skin dismiss it as nothing and avoid going to the doctor until it is too late, such skin cancer apps can save many lives! And indeed, there are already such solutions on the market like that from the Dutch company SkinVision [8]. In addition, AI-powered skin cancer identification apps can be applied in developing countries where only a few dermatologists are available or are located far away from places where patients are living. For example, if in an isolated village at least one smartphone with the corresponding app is available, inhabitants can regularly scan their skin for cancer. In this way, a low-cost universal access to vital skin diagnostic care can be provided in developing regions.

​

Visual identification of diseases is not the only field of medical diagnosis where MI can be applied. A team of Chinese and US researchers have used Machine Learning (ML) for the analysis of diverse and massive amounts of electronic health record data to develop a system that is capable to query electronic health records in a manner similar to the hypothetico-deductive reasoning used by physicians during the diagnosis process [9]. Their system uses ML-based natural language processing techniques to extract clinically relevant information from digital records and was trained on more than 100 million data records of pediatric patients. The system achieved high diagnostic accuracy across multiple organ systems and is comparable to experienced pediatricians in diagnosing common childhood diseases. This and similar systems can support physicians in tackling large amounts of data, augmenting diagnostic evaluations, and providing clinical decision support in cases of diagnostic uncertainty or complexity. Particularly, such systems can be extremely valuable for regions where medical experts are in shortage.

​

​

Drug research

 

The development process for a new medicine is challenging, very time-consuming, and extremely expensive. On an average, it takes 10 to 15 years to develop a new single drug [10]. On the other hand, the average success rate of a drug discovery process is only 10%. These facts lead to extremely high development cost for a single drug with estimated median of $985 million [11]. Altogether it results in a very slow and often unprofitable drug discovery process. Through automated analysis of huge amounts of data and by identifying more promising drug candidates, AI algorithms can help reduce the development costs, increase the success rate, and speed up the development cycle of medicine.

​

And indeed, there are already drugs which were developed with the help of AI: DSP-1181, a molecule of the drug treating obsessive-compulsive disorder, was invented by AI and accepted for a human trial [12]. The development of this drug took only 12 months. Another example of AI usage in drug development is the discovery of six novel inhibitors of the DDR1 gene, a kinase target implicated in fibrosis and other diseases [13]. To achieve this result, the AI system needed only 21 days! The developed substances were tested on mice and showed positive results.

​

Also, during the current COVID-19 pandemic AI is being used heavily for research on potential medicine and vaccines [14]. Indeed, coronavirus has demonstrated an essential need in faster development processes for drugs and vaccines since, according to WHO, it will not be the last pandemic due to worsening ecological crises [15].

​

​

Tackling operational and organizational challenges in healthcare

 

While patient care is the main focus of any healthcare facility, doctors and nurses are often sidetracked from their main activity by the need to handle a lot of administrative tasks and paperwork, e.g., scheduling, claim and bill management, and transferring records from one system to another. These tasks are usually rather simple, digitalized, rule-based, and of high volume – ideal prerequisites to automate these processes using robotic process automation (RPA) (s. e.g., [16] or [17]). RPA is a technology that utilizes “software bots” which can mimic human user interactions with a digital system. It means that an RPA bot is able to perform exactly the same steps on a computer as a doctor or nurse would do them. RPA systems are quite flexible so that any strict rule-based digital process currently performed by a human can be potentially automatized by a bot. Since RPA utilizes software bots, there is no need for any changes in the existing IT landscape of a hospital, thus making the introduction of this innovation very easy and affordable. RPA can also be combined with other MI technologies like AI and ML in order to automize more complex, non-deterministic processes. Altogether, RPA helps free up medical personnel from monotonous paperwork and enables doctors and nurses to focus on what really counts — taking care of the health of their patients.

​

Another MI technique useful for operational excellence in healthcare facilities is Process Mining [18]. During Process Mining, specialized algorithms are applied to the so-called log data — the data that are collected and stored every time an activity in the IT system is performed. These algorithms identify trends, patterns and details contained in the logs and provide insights regarding process flows, process bottlenecks and potential opportunities for optimization. Process mining enables businesses to improve their process efficiency and can also be applied in healthcare facilities. For example, a team of Dutch researchers has applied Process Mining for the analysis of a gynecological oncology process in a Dutch hospital [19] which serves as a proof-of-concept for the application of Process Mining in healthcare facilities.     

 

      

MI healthcare applications in developing countries

 

MI technologies can also be used to enhance healthcare in developing countries. For example, an Indian social startup Wadhwani AI is using AI to tackle tuberculosis. Tuberculosis hotspots are usually identified by income and population density of a region. However, this evaluation approach is very unprecise. Wadhwani AI has developed a system that has learned more granular characteristics of regions at risk [20]. With this approach it is possible to identify potential tuberculosis hotspots also in non-obvious locations. Currently, Wadhwani is also working on an AI-powered system which can diagnose tuberculosis based on cough sounds. Since such a system can be installed on a smartphone, the identification of potential patients can be drastically accelerated with this approach.

​

Another problem that Wadhwani AI is tackling using MI technologies is to identify low-birth-weight babies. Low weight of a new-born is a serious problem; if such a baby does not get appropriate care, low birthweight can result in stunted growth, lower IQ, diabetes, or even early death. Unfortunately, in many regions of the world infants are not weighed at birth or are weighed inaccurately. Wadhwani AI is developing an AI-powered smartphone-based anthropometry tool which creates a 3D model of an infant and estimates all relevant anthropometric parameters. Using this system, frontline volunteers can screen for low-birth-weight babies in underprivileged regions without any specific equipment. This system is currently tested in four states in India across diverse settings like rural homes, primary health facilities, and hospitals.