While it should not be assumed that using a conversation agent will benefit all information system applications, there are some properties associated with it that make them particularly beneficial for this specific application. Rare disease patients find it difficult to get the correct diagnosis within a reasonable period of time and often need to investigate on their own without much technical knowledge.
Numerous studies have shown that these issues are currently not being effectively addressed, with the average time to a correct diagnosis being 4.8 years. By using an informed virtual assistant, accessible 24 hours a day and pleasant, it is possible to support the patient or relative, transmit a sense of confidence, which can have motivating effects, and achieve a diagnosis in much less time. Dr. Rachael is capable of carrying on a normal conversation discussing a variety of topics as well as having a vast amount of knowledge about the symptoms of over 300 rare diseases currently and we are in the process of increasing this number to 1,200 conditions.
Dr. Rachael is not a conventional bot since she is not programmed with questions and answers detecting patterns or specific words, but is based on a general cognitive engine that uses natural language with a vocabulary of more than 50,000 words (in addition to a dictionary of 30,000 scientific words). This cognitive engine acquires knowledge in the form of natural language, files it in the form of ideograms that describe the meaning of each complete sentence and retrieves said knowledge using a form of abduction, association of ideas, eliminating duplication of ideas and with the help of multivariate logic. of first order. To this cognitive engine has been added a module of frequently asked questions and answers (like that of normal bots), abilities such as access to Wikipedia, Wolfram, the weather forecast, and other functionality associated with conventional assistants such as Siri or Cortana. Therefore, Dr. Rachael is perhaps one of the most advanced virtual assistants today.
It is possible to communicate via the keyboard and screen or via speech recognition (using Sphinx) and voice generation with inflection (using maryTTS). When the user makes a comment, Dr. Rachael will make an appropriate related comment, based on an association of ideas much as a person would. If a question posed closely matches a question in the database, Dr. Rachael presents the answer directly to the user. If no such question is found, the closest general comment available is provided. When the user mentions a symptom that Dr. Rachael recognizes, a counter is increased in order to form a small group of conditions that could be associated with the patient's ailment. Each comment most of the time can be matched to more than one condition so that the counters are kept, and when a given statement is matched to more than one condition, all the condition counters are incremented by one. If a given statement is related to only one condition, then only the counter associated with this condition is increased by one. The final rating of the patient's probable condition is calculated based on which counter is the largest, and when there is insufficient confidence of a given diagnosis, the top three conditions are displayed, generating a probability that a patient has been found. similar diagnosis.
At the end of each session, the user is provided with the 3 most likely conditions for which the user or family member is at risk, and a preformulated overview of each condition is provided. This allows the patient or family to get a firm suggestion even if the session lasts a long time or is even completed in many sessions, hopefully helping them in the right direction towards a proper diagnosis in the shortest amount of time. This type of feedback is an integral part of Dr. Rachael's approach, which has been applied previously to diagnosis of eating disorders and personality tests.
