REVIEW PAPER, REVIEW ARTICLE (PREGLEDNI RAD)
Petra Kujundžić1, Tatjana Matijaš1
1 University of Split, University Department of Health Studies, Split, Croatia
Corresponding author: Tatjana Matijaš, University of Split, University Department of Health Studies, Split, Croatia, email:
DOI: https://doi.org/10.55378/rv.47.2.5
Summary
Introduction: Technological progress leads to an increasing use of radiological imaging, and an increase in the number of imaging results in an increased workload for radiologists. The driver of the application of AI in radiology is considered to be the reduction of the workload of radiologists and the need for faster and more accurate diagnosis.
Aim: The aim of this paper is to bring the reader closer to the implementation of AI in radiology, especially in the MRI modality, and how deep learning algorithms improve image reconstruction.
Discussion: Numerous studies have confirmed the importance of implementing machine learning, a subset of artificial intelligence, in the radiology system. In this review paper, numerous researches on the application of deep learning in magnetic resonance imaging are highlighted, and the emphasis is on models for automatic segmentation. Automatic segmentation has shown excellent results in the early detection of osteoarthritis, then in anterior cruciate ligament and meniscus tears, the most common knee injuries, and more recently, the deep learning model has excelled in automatic bone age estimation. Automatic segmentation has achieved, above all, high accuracy and precision, objectivity and time saving.
Conclusion: Previous research has already highlighted the significant advantage of using machine learning in radiology and the exceptional compatibility between the work of radiologists and machine learning, which achieves precise and quick diagnoses. All this is a great incentive for further research, and technological progress will certainly speed up its integration into clinical practice.
Keywords: artificial intelligence; automatic segmentation; deep learning; MRI
Abbreviations and acronyms: AI (Artificial Intelligence), ANN (Artificial Neural Network), CNN (Convolutional Neural Network), DL (Deep Learning), fCNN (Fully Convolutional Neural Network), HNN (Holistically Nested Networks), ML (Machine Learning), MRI (Magnetic Resonance Imaging), NMR (Nuclear Magnetic Resonance), OA (osteoarthritis), SL (Supervised Learning), TBM (Transport Based Morphometry), UL (Unsupervised Learning), VN (Variational Network)
Umjetna inteligencija u oslikavanju koljena magnetnom rezonancijom
Sažetak
Uvod: Tehnološkim napretkom dolazi do sve veće uporabe radioloških snimanja, a povećanjem broja snimanja dolazi do povećanog radnog opterećenja radiologa. Pokretačem primjene AI u radiologiji smatra se upravo smanjenje radnog opterećenja radiologa i potreba za bržom i preciznijom uspostavom dijagnoze.
Cilj rada: Cilj ovog rada je približiti čitatelju implementaciju AI u radiologiji, posebno kod modaliteta MRI te na koji način algoritmi dubokog učenja pospješuju rekonstrukciju slike.
Rasprava: Brojna su istraživanja potvrdila značaj implementacije strojnog učenja, podskupa umjetne inteligencija, u radiološki sustav. U ovom preglednom radu izdvojena su brojna istraživanja primjene dubokog učenja kod magnetne rezonancije, a naglasak je na modelima za automatsku segmentaciju. Automatska segmentacija pokazala je izvrsne rezultate kod ranog otkrivanja osteoartritisa, zatim kod puknuća prednjeg križnog ligamenta i meniska, najčešćih ozljeda koljena, a također se u novije vrijeme model dubokog učenja istaknuo i kod automatske procjene koštane dobi. Automatskom segmentacijom postigla se, prije svega visoka točnost i preciznost, objektivnost i ušteda vremena.
Zaključak: Dosadašnja istraživanja već su istaknula značajnu prednost primjene strojnog učenja u radiologiji te iznimnu kompatibilnost u radu radiologa i strojnog učenja, čime se postižu precizne i brze dijagnoze. Sve je to veliki poticaj za daljnja istraživanja, a tehnološki napredak zasigurno će ubrzati njegovu integraciju u kliničku praksu.
Ključne riječi: automatizirana segmentacija; duboko učenje; MRI; umjetna inteligencija
