Justyna Włodarczyk1, Wadim Wojciechowski2, Kamila Czaplicka3, Andrzej Urbanik4, Zbisław Tabor5. 1. Jagiellonian University, prof. Stanisława Łojasiewicza 11, 30-348 Krakow, Poland. Electronic address: justyna.wlodarczyk@uj.edu.pl. 2. Jagiellonian University Medical College, Kopernika 19, 31-501 Krakow, Poland; Medical Center iMed24, Al. Jana Pawła II 41f, 31-864 Krakow, Poland. Electronic address: wadim@mp.pl. 3. Jagiellonian University, prof. Stanisława Łojasiewicza 11, 30-348 Krakow, Poland. Electronic address: kamila.czaplicka@uj.edu.pl. 4. Jagiellonian University Medical College, Kopernika 19, 31-501 Krakow, Poland. Electronic address: andrzej.urbanik@uj.edu.pl. 5. Cracow University of Technology, Warszawska 24, 31-155 Krakow, Poland. Electronic address: ztabor@pk.edu.pl.
Abstract
PURPOSE: According to current recommendations in diagnostics of rheumatoid arthritis (RA), Magnetic resonance (MR) images of wrist joints are used to evaluate three main signs of RA: synovitis, bone edema and bone erosions. In this paper we present an efficient method for segmentation of 15 bones present on MR images of the wrist which is inevitable for future computer-assisted diagnosis system for RA lesions. METHOD: The segmentation procedure consists of two stages. The first stage is evaluation of markers (parts of bones working as seeds for the watershed algorithm) for bones in every joint: the distal parts of ulna and radius, the proximal parts of metacarpal bones and carpal bones. In the second stage the watershed from markers algorithm is applied based on the markers determined in the previous stage and the wrist bones are segmented. The markers were found using Multi Otsu algorithm along with custom method for filtering bones from other tissues. RESULTS: We analyzed 34 MR images. The automated segmentations were compared with manual segmentations using metrics: accuracy ACC derived from area under ROC curve AUC, Dice coefficient and mean absolute distance MAD. The mean (standard deviation) values of ACC, Dice and MAD were 0.99 (0.02), 0.98 (0.04) and 1.21 (0.39), respectively. CONCLUSION: The results of this study prove that our method is efficient and gives satisfactory results for segmentation of bones on low-field MR images of the wrist.
PURPOSE: According to current recommendations in diagnostics of rheumatoid arthritis (RA), Magnetic resonance (MR) images of wrist joints are used to evaluate three main signs of RA: synovitis, bone edema and bone erosions. In this paper we present an efficient method for segmentation of 15 bones present on MR images of the wrist which is inevitable for future computer-assisted diagnosis system for RA lesions. METHOD: The segmentation procedure consists of two stages. The first stage is evaluation of markers (parts of bones working as seeds for the watershed algorithm) for bones in every joint: the distal parts of ulna and radius, the proximal parts of metacarpal bones and carpal bones. In the second stage the watershed from markers algorithm is applied based on the markers determined in the previous stage and the wrist bones are segmented. The markers were found using Multi Otsu algorithm along with custom method for filtering bones from other tissues. RESULTS: We analyzed 34 MR images. The automated segmentations were compared with manual segmentations using metrics: accuracy ACC derived from area under ROC curve AUC, Dice coefficient and mean absolute distance MAD. The mean (standard deviation) values of ACC, Dice and MAD were 0.99 (0.02), 0.98 (0.04) and 1.21 (0.39), respectively. CONCLUSION: The results of this study prove that our method is efficient and gives satisfactory results for segmentation of bones on low-field MR images of the wrist.
Authors: Lukas Folle; Timo Meinderink; David Simon; Anna-Maria Liphardt; Gerhard Krönke; Georg Schett; Arnd Kleyer; Andreas Maier Journal: Sci Rep Date: 2021-05-06 Impact factor: 4.379