Artificial Intelligence (AI) is a rising field in medicine. It is described as intelligence shown by machines, so one would also call them "intelligent agents". Their behaviour includes cognitive functions, like learning and problem solving [8]. The spread of AI also leads to many different products used in the medical field. These reach from designing patient-specific treatment plans to prediciting the developement of bigger populations (e.g. flu outbreak). All these systems however rely on similar concepts and techniques. In general data is aggregated first, which is then comprehended and analyzed afterwards. To do so, techniques like machine learning, data mining and mathematical algorithms are used.
In the following we will have a look at some examples for the use of AI in medical diagnostics.
IBM Watson Health
IBM Waston is a cognitive system used to create patient-specific treatment plans. Watson is a natural language processing system (NLP). As a first step, Watson has access to different data repositories for different domains, also called corpus. These include unstructured information from the specific field, as scientific articles, clinical trials and available lab data. Hereby Watsons accuracy increases by gathering as much data as possible. By introducing dictionaires and thesauri to the system (nouns, verbs, prepositions,...) Watson learns the domain specific language. Afterwards annotators are used to read and extract specific terms from the scientific literature. By analyzing the available data, Watson can then expand each annotator by similar terms and thereby clustering the information. This learning and the resulting clusters are then visualized by holistic networks. Watson is also able to generate hypotheses for relations from the analyzed data. [3] [4]
In the medical field Watson Health is used for different tasks. Tested in oncology, Watson Health creates a range of treatment plans. To do so, the system analyzes available scientific data as well as the patients medical record. By getting key informations from the medical record, Watson Health is able to create different patient-specific treatment plans. Diagrams and charts are then used to present these analyses to the physician. A study showed that Watson health showed a 90% accordance with medical experts on a tumor board, when both parties designed treatment plans for different tumor patients. [1]
In radiology the Watson health system is used for Clinical Imaging. First it analyzes the available data. Then the clinical report (e.g. medical scans) is compared to the recorded diagnosis. With the available scientific literature, this allows Watson to identify incomplete and incorrect documentations, report them to the physicians and propose possible changes. [2]
Another approach is to generate problem lists from the patients electronic medical report (EMR) and available medical literature. By combining both data pools, Watson is able to mention diseases and disorders mentioned in the EMR and predict possible further outcomes. To limit the available scientific data with regards to the EMR, NLP and machine learning is used. Afterwards Concept Unique Identifiers are produced and concepts are mapped. To better analyze the EMR, different features are evaluated. These include, among others, lexical (how important is the term in the document), structural (where in the record is the term located) and frequency features. [5]
Streams
Streams is a system developed by DeepMind (Google). It's main component is a secure instant app. The system analyzes available medical literature and patients records and includes recent lab results. This is done by using techniques like machine learning and data mining. If the lab results then show critical values that could lead to negative changes in the patients condition, this information is directly forwarded to the right staff member via the app. The alert than also includes a diagnosis based on medical record, scientific literature and lab results. Currently tested in the UK, the system shows great reception around the medical staff. [7]
Other systems
Next to the systems by big, known companies there are also other systems using AI in the medical field, some of which will be described here briefly.
Genome Health Solutions
This system provides a workflow for patients with cancer or other diseases. By analyzing different data sources it predicts when it's the right time to visit the right doctor. Furthermore the system provides patient-specific treatment options. It is planned as a decision-support system. So it doesn't develope one right treatment plan but rather developes several ones and allows the physician to choose between these options. [6]
Predixion Software
Here data is again analyzed from different sources using machine learning, data mining and other algorithms. The main goal is to prevent readmission of patients. Therefore the patients data and other data sources are analyzed during a hospital stay. From this analysis the best time to release the patient is predicted. [6]
Practice Fusion
This is a free, cloud-based EMR platform. Data from the EMR and other public data sources is analyzed. Based on this analysis, treatment plans are developed and the health can be monitored. Furthermore the system allows to analyze and predict health changes in bigger populations by analyzing data from different sources. [6]
To look at a short instruction of some more systems have a look here.