RESUMEN
The aim of the study is to evaluate the efficacy of approaches to sampling during periodic quality control of the artificial intelligence (AI) results in biomedical practice. Materials and Methods: The approaches to sampling based on point statistical estimation, statistical hypothesis testing, employing ready-made statistical tables, as well as options of the approaches presented in GOST R ISO 2859-1-2007 "Statistical methods. Sampling procedures for inspection by attributes" have been analyzed. We have considered variants of sampling of different sizes for general populations from 1000 to 100,000 studies.The analysis of the approaches to sampling was carried out as part of an experiment on the use of innovative technologies in computer vision for the analysis of medical images and their further application in the healthcare system of Moscow (Russia). Results: Ready-made tables have specific statistical input data, which does not make them a universal option for biomedical research. Point statistical estimation helps to calculate a sample based on given statistical parameters with a certain confidence interval. This approach is promising in the case when only a type I error is important for the researcher, and a type II error is not a priority. Using the approach based on statistical hypothesis testing makes it possible to take account of type I and II errors based on the given statistical parameters. The application of GOST R ISO 2859-1-2007 for sampling allows using ready-made values depending on the given statistical parameters.When evaluating the efficacy of the studied approaches, it was found that for our purposes, the optimal number of studies during AI quality control for the analysis of medical images is 80 items. This meets the requirements of representativeness, balance of the risks to the consumer and the AI service provider, as well as optimization of labor costs of employees involved in the process of quality control of the AI results.
Asunto(s)
Investigación Biomédica , Trabajo de Parto , Humanos , Embarazo , Femenino , Inteligencia Artificial , Control de Calidad , Diseño Interior y MobiliarioRESUMEN
The aim of the study was to develop a methodology for conducting post-registration clinical monitoring of software as a medical device based on artificial intelligence technologies (SaMD-AI). Materials and Methods: The methodology of post-registration clinical monitoring is based on the requirements of regulatory legal acts issued by the Board of the Eurasian Economic Commission. To comply with these requirements, the monitoring involves submission of the review of adverse events reports, the review of developers' routine reports on the safety and efficiency of SaMD-AI, and the assessment of the system for collecting and analyzing developers' post-registration data on the safety and efficiency of medical devices. The methodology was developed with regard to the recommendations of the International Medical Device Regulators Forum and the documents issued by the Food and Drug Administration (USA). Field-testing of this methodology was carried out using SaMD-AI designed for diagnostic imaging. Results: The post-registration monitoring of SaMD-AI consists of three key stages: collecting user feedback, technical monitoring and clinical validation. Technical monitoring involves routine evaluation of SaMD-AI output data quality to detect and remove flaws in a timely manner, and to secure the product stability. Major outcomes include an ordered list of technical flaws in SaMD-AI and their classification using evidence from diagnostic imaging studies. The application of this methodology resulted in a gradual reduction in the number of studies with flaws due to timely improvements in artificial intelligence algorithms: the number of flaws decreased to 5% in various aspects during subsequent testing. Clinical validation confirmed that SaMD-AI is capable of producing clinically meaningful outputs related to its intended use within the functionality determined by the developer. The testing procedure and the baseline testing framework were established during the field testing. Conclusion: The developed methodology will ensure the safety and efficiency of SaMD-AI taking into account its specifics as intangible medical devices. The methodology presented in this paper can be used by SaMD-AI developers to plan and carry out the post-registration clinical monitoring.
Asunto(s)
Inteligencia Artificial , Programas Informáticos , Estados Unidos , Algoritmos , Vigilancia de Productos ComercializadosRESUMEN
BACKGROUND AND OBJECTIVE: Lung cancer is the most common type of cancer with a high mortality rate. Early detection using medical imaging is critically important for the long-term survival of the patients. Computer-aided diagnosis (CAD) tools can potentially reduce the number of incorrect interpretations of medical image data by radiologists. Datasets with adequate sample size, annotation, and truth are the dominant factors in developing and training effective CAD algorithms. The objective of this study was to produce a practical approach and a tool for the creation of medical image datasets. METHODS: The proposed model uses the modified maximum transverse diameter approach to mark a putative lung nodule. The modification involves the possibility to use a set of overlapping spheres of appropriate size to approximate the shape of the nodule. The algorithm embedded in the model also groups the marks made by different readers for the same lesion. We used the data of 536 randomly selected patients of Moscow outpatient clinics to create a dataset of standard-dose chest computed tomography (CT) scans utilizing the double-reading approach with arbitration. Six volunteer radiologists independently produced a report for each scan using the proposed model with the main focus on the detection of lesions with sizes ranging from 3 to 30 mm. After this, an arbitrator reviewed their marks and annotations. RESULTS: The maximum transverse diameter approach outperformed the alternative methods (3D box, ellipsoid, and complete outline construction) in a study of 10,000 computer-generated tumor models of different shapes in terms of accuracy and speed of nodule shape approximation. The markup and annotation of the CTLungCa-500 dataset revealed 72 studies containing no lung nodules. The remaining 464 CT scans contained 3151 lesions marked by at least one radiologist: 56%, 14%, and 29% of the lesions were malignant, benign, and non-nodular, respectively. 2887 lesions have the target size of 3-30 mm. Only 70 nodules were uniformly identified by all the six readers. An increase in the number of independent readers providing CT scans interpretations led to an accuracy increase associated with a decrease in agreement. The dataset markup process took three working weeks. CONCLUSIONS: The developed cluster model simplifies the collaborative and crowdsourced creation of image repositories and makes it time-efficient. Our proof-of-concept dataset provides a valuable source of annotated medical imaging data for training CAD algorithms aimed at early detection of lung nodules. The tool and the dataset are publicly available at https://github.com/Center-of-Diagnostics-and-Telemedicine/FAnTom.git and https://mosmed.ai/en/datasets/ct_lungcancer_500/, respectively.