Design criteria for AI-based IT systems.

PURPOSE: This editorial relates to a panel discussion during the CARS 2023 congress that addressed the question on how AI-based IT systems should be designed that record and (transparently) display a reproducible path on clinical decision making. Even though the software engineering approach suggested for this endeavor is of a generic nature, it is assumed that the listed design criteria are applicable to IT system development also for the domain of radiology and surgery. METHODS: An example of a possible design approach is outlined by illustrating on how to move from data, information, knowledge and models to wisdom-based decision making in the context of a conceptual GPT system design. In all these design steps, the essential requirements for system quality, information quality, and service quality may be realized by following the design cycle as suggested by A.R. Hevner, appropriately applied to AI-based IT systems design. RESULTS: It can be observed that certain state-of-the-art AI algorithms and systems, such as large language models or generative pre-trained transformers (GPTs), are becoming increasingly complex and, therefore, need to be rigorously examined to render them transparent and comprehensible in their usage for all stakeholders involved in health care. Further critical questions that need to be addressed are outlined and complemented with some suggestions, that a possible design framework for a stakeholder specific AI system could be a (modest) GPT based on a small language model. DISCUSSION: A fundamental question for the future remains whether society wants a quasi-wisdom-oriented healthcare system, based on data-driven intelligence with AI, or a human curated wisdom based on model-driven intelligence (with and without AI). Special CARS workshops and think tanks are planned to address this challenging question and possible new direction for assisting selected medical disciplines, e.g., radiology and surgery.

The clinical day at CARS: philosophy, strategy, and the future.

PURPOSE: For over three decades, the Computer-Assisted Radiology and Surgery (CARS) International Congress and Exhibition has provided a forum for the presentation of innovations in computer applications in radiology, medicine, and surgery. A unique feature of the CARS meetings is the interplay between scientists, engineers, and physicians. Since 2007, a Clinical Day program was introduced to the Congress highlighting the practical applications of new technology within the context of clinical medical and surgical practice. METHODS: The Clinical Day of the CARS Congress allows cross-fertilization of ideas between technologically oriented engineers and clinically oriented physicians; two groups who typically have little interaction. Activities of the Clinical Day include presentations by invited speakers, presentations of Innovative Clinical Investigations, a Panel Discussion and Open Forum, and, most recently, real-time clinical presentations with professionally prepared scholarly videos. Special consideration is given to young researchers and students. There has been an explosion of interest in the Clinical Day with continued and growing interest in Artificial Intelligence, Computer-Assisted Surgery, and new scientific breakthroughs as they become linked to clinical applications. RESULTS: Success of the Clinical Day is emphasized by increased participation and efforts to expand the scope and depth of Clinical Day activities. The Open Forum has proven to be highly effective in identifying important new technologic challenges in medicine and promoting discussion among those whose expertise likely can lead to solutions. CONCLUSIONS: The original goal of the Clinical Day, to provide an effective means to "bridge the gap" between the engineering community and practicing physicians and surgeons, has been realized through the presentation and discussion of real-life, clinical material that utilizes advanced technology. The program has served to inspire young researchers by allowing them to see the end results of their laboratory investigations, thereby gaining a greater appreciation of the importance of their work.

Validation workflow for a clinical Bayesian network model in multidisciplinary decision making in head and neck oncology treatment.

PURPOSE: Oncological treatment is being increasingly complex, and therefore, decision making in multidisciplinary teams is becoming the key activity in the clinical pathways. The increased complexity is related to the number and variability of possible treatment decisions that may be relevant to a patient. In this paper, we describe validation of a multidisciplinary cancer treatment decision in the clinical domain of head and neck oncology. METHOD: Probabilistic graphical models and corresponding inference algorithms, in the form of Bayesian networks, can support complex decision-making processes by providing a mathematically reproducible and transparent advice. The quality of BN-based advice depends on the quality of the model. Therefore, it is vital to validate the model before it is applied in practice. RESULTS: For an example BN subnetwork of laryngeal cancer with 303 variables, we evaluated 66 patient records. To validate the model on this dataset, a validation workflow was applied in combination with quantitative and qualitative analyses. In the subsequent analyses, we observed four sources of imprecise predictions: incorrect data, incomplete patient data, outvoting relevant observations, and incorrect model. Finally, the four problems were solved by modifying the data and the model. CONCLUSION: The presented validation effort is related to the model complexity. For simpler models, the validation workflow is the same, although it may require fewer validation methods. The validation success is related to the model's well-founded knowledge base. The remaining laryngeal cancer model may disclose additional sources of imprecise predictions.

A Guide for Constructing Bayesian Network Graphs of Cancer Treatment Decisions.

In complex cancer cases, Bayesian networks can support clinical experts in finding the best patient-specific therapeutic decisions. However, the development of decision networks requires teamwork of at least one domain expert and one knowledge engineer making the process expensive, time-consuming, and prone to misunderstandings. We present a novel method for guided modeling. This method enables domain experts to model collaboratively without the need of knowledge engineers, increasing both the development speed and model quality.

Accessing Patient Information for Probabilistic Patient Models Using Existing Standards.

Clinical decision support systems (CDSS) are developed to facilitate physicians' decision making, particularly for complex, oncological diseases. Access to relevant patient specific information from electronic health records (EHR) is limited to the structure and transmission formats in the respective hospital information system. We propose a system-architecture for a standardized access to patient specific information for a CDSS for laryngeal cancer. Following the idea of a CDSS using Bayesian Networks, we developed an architecture concept applying clinical standards. We recommend the application of Arden Syntax for the definition and processing of needed medical knowledge and clinical information, as well as the use of HL7 FHIR to identify the relevant data elements in an EHR to increase the interoperability the CDSS.

Model-guided therapy for hepatocellular carcinoma: a role for information technology in predictive, preventive and personalized medicine.

Predictive, preventive and personalized medicine (PPPM) may have the potential to eventually improve the nature of health care delivery. However, the tools required for a practical and comprehensive form of PPPM that is capable of handling the vast amounts of medical information that is currently available are currently lacking. This article reviews a rationale and method for combining and integrating diagnostic and therapeutic management with information technology (IT), in a manner that supports patients through their continuum of care. It is imperative that any program devised to explore and develop personalized health care delivery must be firmly rooted in clinically confirmed and accepted principles and technologies. Therefore, a use case, relating to hepatocellular carcinoma (HCC), was developed. The approach to the management of medical information we have taken is based on model theory and seeks to implement a form of model-guided therapy (MGT) that can be used as a decision support system in the treatment of patients with HCC. The IT structures to be utilized in MGT include a therapy imaging and model management system (TIMMS) and a digital patient model (DPM). The system that we propose will utilize patient modeling techniques to generate valid DPMs (which factor in age, physiologic condition, disease and co-morbidities, genetics, biomarkers and responses to previous treatments). We may, then, be able to develop a statistically valid methodology, on an individual basis, to predict certain diseases or conditions, to predict certain treatment outcomes, to prevent certain diseases or complications and to develop treatment regimens that are personalized for that particular patient. An IT system for predictive, preventive and personalized medicine (ITS-PM) for HCC is presented to provide a comprehensive system to provide unified access to general medical and patient-specific information for medical researchers and health care providers from different disciplines including hepatologists, gastroenterologists, medical and surgical oncologists, liver transplant teams, interventional radiologists and radiation oncologists. The article concludes with a review providing an outlook and recommendations for the application of MGT to enhance the medical management of HCC through PPPM.

Towards personal health care with model-guided medicine: long-term PPPM-related strategies and realisation opportunities within 'Horizon 2020'.

At the international EPMA Summit carried out in the EU Parliament (September 2013), the main challenges in Predictive, Preventive and Personalised Medicine have been discussed and strategies outlined in order to implement scientific and technological innovation in medicine and healthcare utilising new strategic programmes such as 'Horizon 2020'. The joint EPMA (European Association for Predictive, Preventive and Personalised Medicine) / IFCARS (International Foundation for Computer Assisted Radiology and Surgery) paper emphasises the consolidate position of the leading experts who are aware of the great responsibility of being on a forefront of predictive, preventive and personalised medicine. Both societies consider long-term international partnerships and multidisciplinary projects to create PPPM relevant innovation in science, technological tools and practical implementation in healthcare. Personalisation in healthcare urgently needs innovation in design of PPPM-related medical services, new products, research, education, didactic materials, propagation of targeted prevention in the society and treatments tailored to the person. For the paradigm shift from delayed reactive to predictive, preventive and personalised medicine, a new culture should be created in communication between individual professional domains, between doctor and patient, as well as in communication with individual social (sub)groups and patient cohorts. This is a long-term mission in personalised healthcare with the whole spectrum of instruments available and to be created in the field.

3D++ Visualisation of MEBN Graphs and Screen Representations of Patient Models (PIXIE II).

The aim of this study is to investigate modelling- and visualisation methods and tools for transparent, reproducible and comprehensible information management of patient probabilistic graphical models such as Multi-Entity Bayesian Networks (MEBN). In therapy planning environments, models provide the knowledge base to assist physicians in their decision making process and are typically at the heart of a networked information system. The topic of user interface design (UID) and specifically GUIs needs to be addressed in this context in a very creative manner, if the aim is to make complex models easier for the user to understand and to manage. As a basis for preparing a demonstration to show a proof of concept for the visualisation of MEBNs, the visualization tool called "ANTz" is being proposed here and used as the preferred visualisation tool.

Knowledge engineering for health: a new discipline required to bridge the "ICT gap" between research and healthcare.

Despite vast amount of money and research being channeled toward biomedical research, relatively little impact has been made on routine clinical practice. At the heart of this failure is the information and communication technology "chasm" that exists between research and healthcare. A new focus on "knowledge engineering for health" is needed to facilitate knowledge transmission across the research-healthcare gap. This discipline is required to engineer the bidirectional flow of data: processing research data and knowledge to identify clinically relevant advances and delivering these into healthcare use; conversely, making outcomes from the practice of medicine suitably available for use by the research community. This system will be able to self-optimize in that outcomes for patients treated by decisions that were based on the latest research knowledge will be fed back to the research world. A series of meetings, culminating in the "I-Health 2011" workshop, have brought together interdisciplinary experts to map the challenges and requirements for such a system. Here, we describe the main conclusions from these meetings. An "I4Health" interdisciplinary network of experts now exists to promote the key aims and objectives, namely "integrating and interpreting information for individualized healthcare," by developing the "knowledge engineering for health" domain.

Prediction of intraoperative complexity from preoperative patient data for laparoscopic cholecystectomy.

OBJECTIVE: Different reasons may cause difficult intraoperative surgical situations. This study aims to predict intraoperative complexity by classifying and evaluating preoperative patient data. The basic prediction problem addressed in this paper involves the classification of preoperative data into two classes: easy (Class 0) and complex (Class 1) surgeries. METHODS AND MATERIAL: preoperative patient data were collected from 337 patients admitted to the Klinikum rechts der Isar hospital in Munich, Germany for laparoscopic cholecystectomy (LAPCHOL) in the period of 2005-2008. The data include the patient's body mass index (BMI), sex, inflammation, wall thickening, age and history of previous surgery, as well as the name and level of experience of the operating surgeon. The operating surgeon was asked to label the intraoperative complexity after the surgery: '0' if the surgery was easy and '1' if it was complex. For the classification task a set of classifiers was evaluated, including linear discriminant classifier (LDC), quadratic discriminant classifier (QDC), Parzen and support vector machine (SVM). Moreover, feature-selection was applied to derive the optimal preoperative patient parameters for predicting intraoperative complexity. RESULTS: Classification results indicate a preference for the LDC in terms of classification error, although the SVM classifier is preferred in terms of results concerning the area under the curve. The trained LDC or SVM classifier can therefore be used in preoperative settings to predict complexity from preoperative patient data with classification error rates below 17%. Moreover, feature-selection results identify bias in the process of labelling surgical complexity, although this bias is irrelevant for patients with inflammation, wall thickening, male sex and high BMI. These patients tend to be at high risk for complex LAPCHOL surgeries, regardless of labelling bias. CONCLUSIONS: Intraoperative complexity can be predicted before surgery according to preoperative data with accuracy up to 83% using an LDC or SVM classifier. The set of features that are relevant for predicting complexity includes inflammation, wall thickening, sex and BMI score.

Short history of PACS (Part II: Europe).

Although the concept of picture archiving and communications systems (PACS) was developed in Europe during the latter part of the 1970s, no working system was completed at that time. The first PACS implementations took place in the United States in the early 1980s, e.g. at Pennsylvania University, UCLA, and Kansas City University. Some more or less successful PACS developments also took place in Europe in the 1980s, particularly in the Netherlands, Belgium, Austria, the United Kingdom, France, Italy, Scandinavia, and Germany. Most systems could be characterized by their focus on a single department, such as radiology or nuclear medicine. European hospital-wide PACS with high visibility evolved in the early 1990s in London (Hammersmith Hospital) and Vienna (SMZO). These were followed during the latter part of the 1990s by approximately 10-20 PACS installations in each of the major industrialized countries of Europe. Wide-area PACS covering several health care institutions in a region are now in the process of being implemented in a number of European countries. Because of limitations of space some countries, for example, Denmark, Finland, Spain, Greece, as well as Eastern European countries, etc. could not be appropriately represented in this paper.