Open AccessValidation of an automatic segmentation method to detect vertebral interfaces in ultrasound images
Author(s) -
Aventaggiato Matteo,
Conversano Francesco,
Pisani Paola,
Casciaro Ernesto,
Franchini Roberto,
LayEkuakille Aimè,
Muratore Maurizio,
Casciaro Sergio
Publication year - 2016
Publication title -
iet science, measurement and technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.418
H-Index - 49
eISSN - 1751-8830
pISSN - 1751-8822
DOI - 10.1049/iet-smt.2015.0042
Subject(s) - computer science , segmentation , artificial intelligence , ultrasound , computer vision , radiology , medicine
Aim of this study was to perform a detailed clinical validation of a new fully automatic algorithm for vertebral interface segmentation in echographic images. Abdominal echographic scans of lumbar vertebrae L1–L4 were carried out on 150 female subjects with variable age and body mass index (BMI). Acquired datasets were automatically processed by the algorithm and the accuracy of the obtained segmentations was then evaluated by three independent experienced operators. Obtained results showed a very good specificity in vertebra detection (93.3%), coupled with a reasonable sensitivity (68.1%), representing a suitable compromise between the detection of a sufficient number of vertebrae for reliable diagnoses and the limitation of the corresponding computation time. Importantly, there was only a minimum presence of ‘false vertebrae’ detected (2.8%), resulting in a very low influence on subsequent diagnostic analyses. Furthermore, the algorithm was specifically tuned to provide an improved sensitivity (up to 73.1%) with increasing patient BMI, to keep a suitable number of correctly detected vertebrae even when the acquisition was intrinsically more difficult because of the augmented thickness of abdominal soft tissues. The proposed algorithm will represent an essential added value for developing echographic methods for the diagnosis of osteoporosis on lumbar vertebrae.