Abstract in Englisch:

The primary concern of the available Medical Diagnosis Systems (MDSs), including the state of the art, is to find the perfect link between the patient’s medical record and their health knowledge. As a result, regardless of how powerful or ground-breaking they are in performing this action, it is always possible that their final strong deduction is based on some incomplete input, which may very likely lead to a misdiagnosis. To date, no computer-aided Decision Support System (DSS) has been introduced to address this issue and this is the reason why the available MDSs cannot be well integrated into the workflows of the healthcare institutions. Prior to using these systems, a physician should complete the patient’s medical record by performing a complaint-directed History and Physical examination (H&P) and prepare the required thorough input. The H&P is steered by Differential Diagnosis (DDx), which is the process of differentiating between two or more conditions which share similar signs or symptoms. When a physician performs this examination to provide an MDS with the right input, if no complications occur, at the end of the H&P s(he) will reach the diagnosis too, and as a result the later use of the MDS would be of less value. Consequently, MDSs are used very seldom in practice and are exclusively used in complicated medical cases, where the H&P performed by the physician has not led to a definitive diagnosis. This study aims to introduce a Diagnostic Decision Support System (DDSS) that guides the user in performing DDx directed H&Ps. Of course, this system can be used by the experts who are originally meant to perform the H&P, i.e., physicians and if allowed by the healthcare system of the country Nurse Practitioners (NPs) and/or Physician Assistances (PAs), in order to have some reminders while completing the H&P report. However, as the shortage of medical doctors is worsening in the recent years, this system intends to guide simple nurses in performing the H&P, helping the doctors to be able to see more patients in a certain period of time, as they would just need to review and asses the already prepared H&P report. In third world countries with medical treatment being far away, the user of the system can even be a person with some basic medical knowledge or the patient him- or herself, who may use the system in order to become aware of the possibilities and receive suggestions on finding the right experts to be contacted. When used by experts, as might be expected, in case of any complications it is again possible to use the output of this system as the input of available MDSs that are designed to address such cases. The differential diagnostic problem can be recursively broken down into sub-problems by weighting the likelihood of the presence of possible diseases. These subproblems may induce different abstraction levels and can be of different granularities. Moreover, according to the nature of DDx, the problem solvers should be collaborative and those dealing with similar diseases need to have more communications, which are to be conducted in a timely manner. These features clearly show that the DDx domain meets the characteristics of the holonic domains, which involve agents that can simultaneously be a whole and a part. Con-sequently, a Holonic Multi-Agent System (HMAS) composed of agents that are either expert in diagnosing a disease or a group of related diseases is proposed in this study to address the DDx problem. From the complexity of the medical data the system must deal with and the fact that medical knowledge demonstrates a steady upward growth it can be inferred that the proposed system should be capable of learning and adaptation to the new findings in this field. Moreover, since diagnosis is very much affected by the geographical regions, the system should also be capable of learning and adaptation to the local patterns too. This means that the proposed system should be empowered with appropriate Machine Learning (ML) techniques. As the holonic approach is a new trend in computer science there are very few studies on the learning techniques that can be applied to HMASs. Accordingly, this research also includes a study on determining the right ML techniques for the objectives of the system. Therefore, even though the development of the Holonic Medical Diagnosis System (HMDS), which is capable of performing DDx, is the practical contribution of this work, the introduction of the ML techniques that are used to adapt its functionality can be considered as the conceptual/theoretical contribution of this research, as the proposed techniques can be applied to other HMASs that adopt a similar approach for problem solving to the one followed in this study. This work also includes assessment simulations of the proposed system that monitor the system’s general behavior in performing the H&P and examine the learning abilities of the system by providing the system with appropriate inputs and evaluating the corresponding outputs. The results of these assessments show that the proposed system is a promising tool for addressing the DDx problem and eventually helping the MDSs to gain acceptance from the health service providers in practice.