Search

In this paper, we study human–AI collaboration protocols, a design-oriented construct aimed at establishing and evaluating how humans and AI can collaborate in cognitive tasks. We applied this construct in two user studies involving 12 specialist radiologists (the knee MRI study) and 44 ECG readers of varying expertise (the ECG study), who evaluated 240 and 20 cases, respectively, in different collaboration configurations. We confirm the utility of AI support but find that XAI can be associated with a "white-box paradox", producing a null or detrimental effect. We also find that the order of presentation matters: AI-first protocols are associated with higher diagnostic accuracy than human-first protocols, and with higher accuracy than both humans and AI alone. Our findings identify the best conditions for AI to augment human diagnostic skills, rather than trigger dysfunctional responses and cognitive biases that can undermine decision effectiveness. • We study how humans and AI can collaborate in cognitive tasks in the medical setting. • Two studies involved 12 radiologists and 44 ECG readers (for 240 and 20 cases, resp.) • AI support was found useful but XAI was associated with a null or detrimental effect. • AI-first protocols had higher accuracy than human-first ones and humans or AI alone. • Our findings identify the best conditions for AI to augment human diagnostic skills. [ABSTRACT FROM AUTHOR]

Objective: Proper diagnosis of Low Back Pain (LBP) is quite challenging in especially the developing countries like India. Though some developed countries prepared guidelines for evaluation of LBP with tests to detect psychological overlay, implementation of the recommendations becomes quite difficult in regular clinical practice, and different specialties of medicine offer different modes of management. Aiming at offering an expert-level diagnosis for the patients having LBP, this paper uses Artificial Intelligence (AI) to derive a clinically justified and highly sensitive LBP resolution technique.Materials and Methods: The paper considers exhaustive knowledge for different LBP disorders (classified based on different pain generators), which have been represented using lattice structures to ensure completeness, non-redundancy, and optimality in the design of knowledge base. Further the representational enhancement of the knowledge has been done through construction of a hierarchical network, called RuleNet, using the concept of partially-ordered set (poset) with respect to the subset equality (⊆) relation. With implicit incorporation of probability within the knowledge, the RuleNet is used to derive reliable resolution logic along with effective resolution of uncertainties during clinical decision making.Results: The proposed methodology has been validated with clinical records of seventy seven LBP patients accessed from the database of ESI Hospital Sealdah, India over a period of one year from 2018 to 2019. Achieving 83% sensitivity of the proposed technique, the pain experts at the hospital find the design clinically satisfactory. The inferred outcomes have also been found to be homogeneous with the actual or original diagnosis.Discussions: The proposed approach achieves the clinical and computational efficiency by limiting the shortcomings of the existing methodologies for AI-based LBP diagnosis. While computational efficiency (with respect to both time and space complexity) is ensured by inferring clinical decisions through optimal processing of the knowledge items using poset, the clinical acceptability has been ascertained reaching to the most-likely diagnostic outcomes through probabilistic resolution of clinical uncertainties.Conclusion: The derived resolution technique, when embedded in LBP medical expert systems, would provide a fast, reliable, and affordable healthcare solution for this ailment to a wider range of general population suffering from LBP. The proposed scheme would significantly reduce the controversies and confusion in LBP treatment, and cut down the cost of unnecessary or inappropriate treatment and referral. [ABSTRACT FROM AUTHOR]

Resembling the role of disease diagnosis in Western medicine, pathogenesis (also called Bing Ji) diagnosis is one of the utmost important tasks in traditional Chinese medicine (TCM). In TCM theory, pathogenesis is a complex system composed of a group of interrelated factors, which is highly consistent with the character of systems science (SS). In this paper, we introduce a heuristic definition called pathogenesis network (PN) to represent pathogenesis in the form of the directed graph. Accordingly, a computational method of pathogenesis diagnosis, called network differentiation (ND), is proposed by integrating the holism principle in SS. ND consists of three stages. The first stage is to generate all possible diagnoses by Cartesian Product operated on specified prior knowledge corresponding to the input symptoms. The second stage is to screen the validated diagnoses by holism principle. The third stage is to pick out the clinical diagnosis by physician-computer interaction. Some theorems are stated and proved for the further optimization of ND in this paper. We conducted simulation experiments on 100 clinical cases. The experimental results show that our proposed method has an excellent capability to fit the holistic thinking in the process of physician inference. [ABSTRACT FROM AUTHOR]

Artificial intelligence is a broad field that comprises a wide range of techniques, where deep learning is presently the one with the most impact. Moreover, the medical field is an area where data both complex and massive and the importance of the decisions made by doctors make it one of the fields in which deep learning techniques can have the greatest impact. A systematic review following the Cochrane recommendations with a multidisciplinary team comprised of physicians, research methodologists and computer scientists has been conducted. This survey aims to identify the main therapeutic areas and the deep learning models used for diagnosis and treatment tasks. The most relevant databases included were MedLine, Embase, Cochrane Central, Astrophysics Data System, Europe PubMed Central, Web of Science and Science Direct. An inclusion and exclusion criteria were defined and applied in the first and second peer review screening. A set of quality criteria was developed to select the papers obtained after the second screening. Finally, 126 studies from the initial 3493 papers were selected and 64 were described. Results show that the number of publications on deep learning in medicine is increasing every year. Also, convolutional neural networks are the most widely used models and the most developed area is oncology where they are used mainly for image analysis. [ABSTRACT FROM AUTHOR]