Background: Chest X-ray (CXR) remains the standard imaging modality in postoperative care after non-cardiac thoracic surgery. Lung ultrasound (LUS) showed promising results in CXR reduction. The aim of this review was to identify areas where the evaluation of LUS videos by artificial intelligence could improve the implementation of LUS in thoracic surgery. Methods: A literature review of the replacement of the CXR by LUS after thoracic surgery and the evaluation of LUS videos by artificial intelligence after thoracic surgery was conducted in Medline. Results: Here, eight out of 10 reviewed studies evaluating LUS in CXR reduction showed that LUS can reduce CXR without a negative impact on patient outcome after thoracic surgery. No studies on the evaluation of LUS signs by artificial intelligence after thoracic surgery were found. Conclusion: LUS can reduce CXR after thoracic surgery. We presume that artificial intelligence could help increase the LUS accuracy, objectify the LUS findings, shorten the learning curve, and decrease the number of inconclusive results. To confirm this assumption, clinical trials are necessary. This research is funded by the Slovak Research and Development Agency, grant number APVV 20-0232.

Data streams can be defined as the continuous stream of data in many forms coming from different sources. Data streams are usually non-stationary with continually changing their underlying structure. Solving of predictive or classification tasks on such data must consider this aspect. Traditional machine learning models applied on the drifting data may become invalid in the case when a concept change appears. To tackle this problem, we must utilize special adaptive learning models, which utilize various tools able to reflect the drifting data. One of the most popular groups of such methods are adaptive ensembles. This paper describes the work focused on the design and implementation of a novel adaptive ensemble learning model, which is based on the construction of a robust ensemble consisting of a heterogeneous set of its members. We used k-NN, Naive Bayes and Hoeffding trees as base learners and implemented an update mechanism, which considers dynamic class-weighting and Q statistics diversity calculation to ensure the diversity of the ensemble. The model was experimentally evaluated on the streaming datasets, and the effects of the diversity calculation were analyzed.

Deep neural network models have produced significant results in solving various challenging tasks, including medical diagnostics. To increase the credibility of these black-box models in the eyes of doctors, it is necessary to focus on their explainability. Several papers have been published combining deep learning methods with selected types of explainability methods, usually aimed at analyzing medical image data, including ECG images. The ECG is specific because its image representation is only a secondary visualization of stream data from sensors. However, explainability methods for stream data are rarely investigated. Therefore, in this article we focus on the explainability of black-box models for stream data from 12-lead ECG. We designed and implemented a perturbation explainability method and verified it in a user study on a group of medical students with experience in ECG tagging in their final years of study. The results demonstrate the suitability of the proposed method, as well as the importance of including multiple data sources in the diagnostic process.

Data in many application domains provide a valuable source for analysis and data-driven decision support. On the other hand, legislative restrictions are provided, especially on personal data and patients’ data in the medical domain. In order to maximize the use of data for decision purposes and comply with legislation, sensitive data needs to be properly anonymized or synthetized. This article contributes to the area of medical records synthesis. We first introduce this topic and present it in a broader context, as well as in terms of methods used and metrics for their evaluation. Based on the related work analysis, we selected CTGAN neural network model for data synthesis and experimentally validated it on three different medical datasets. The results were evaluated both quantitatively by means of selected metrics as well as qualitatively by means of proper visualization techniques. The results showed that in most cases, the synthesized dataset is a very good approximation of the original one, with similar prediction performance.