Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry

Abstract

Thyroid volumetry is crucial in the diagnosis, treatment, and monitoring of thyroid diseases. However, conventional thyroid volumetry with 2D ultrasound is highly operator-dependent. This study compares 2D and tracked 3D ultrasound with an automatic thyroid segmentation based on a deep neural network regarding inter- and intraobserver variability, time, and accuracy. Volume reference was MRI. 28 healthy volunteers (24—50 a) were scanned with 2D and 3D ultrasound (and by MRI) by three physicians (MD 1, 2, 3) with different experience levels (6, 4, and 1 a). In the 2D scans, the thyroid lobe volumes were calculated with the ellipsoid formula. A convolutional deep neural network (CNN) automatically segmented the 3D thyroid lobes. 26, 6, and 6 random lobe scans were used for training, validation, and testing, respectively. On MRI (T1 VIBE sequence) the thyroid was manually segmented by an experienced MD. MRI thyroid volumes ranged from 2.8 to 16.7ml (mean 7.4, SD 3.05). The CNN was trained to obtain an average Dice score of 0.94. The interobserver variability comparing two MDs showed mean differences for 2D and 3D respectively of 0.58 to 0.52ml (MD1 vs. 2), −1.33 to −0.17ml (MD1 vs. 3) and −1.89 to −0.70ml (MD2 vs. 3). Paired samples t-tests showed significant differences for 2D (p = .140, p = .002 and p = .002) and none for 3D (p = .176, p = .722 and p = .057). Intraobsever variability was similar for 2D and 3D ultrasound. Comparison of ultrasound volumes and MRI volumes showed a significant difference for the 2D volumetry of all MDs (p = .002, p = .009, p <.001), and no significant difference for 3D ultrasound (p = .292, p = .686, p = 0.091). Acquisition time was significantly shorter for 3D ultrasound. Tracked 3D ultrasound combined with a CNN segmentation significantly reduces interobserver variability in thyroid volumetry and increases the accuracy of the measurements with shorter acquisition times.

Artificial Intelligence-powered automatic volume calculation in medical images – available tools, performance and challenges for nuclear medicine

“…Krönke et al [55] presented a comparison between thyroid volumetry based on 2D US and 3D US in 28 healthy volunteers. They also acquired MR (T1 VIBE) to compute the reference volume of the thyroid.…”

“…Artificial neural networks can efficiently process 3D images, segment structures of interest, and based on those segmentations, calculate volumes with high accuracy. Some AI techniques are capable of adapting to variations in image quality and anatomical structures, improving consistency and reproducibility in volume measurements across different observers and imaging sessions compared to humans in several studies (e. g., [2][3][4]). The major impact of AI can be shown in 3D ultrasonography, a growing field given the drop in costs (due to the possibility of reimbursement and the availability of cheaper devices), where the steep learning curve currently makes it difficult for users to exploit its full potential [2,5].…”

Standardisierte Akquisition und Dokumentation von Videosequenzen bei der konventionellen Schilddrüsensonografie

“…Neben einem speicherbaren 3D-DICOM-Datensatz werden das Schilddrüsenvolumen sowie die Lokalisation, Größe und ACR-TI-RADS-Klassifikation von Knoten ausgegeben. Wie zuverlässig das System (auch bei komplexen Pathologien) funktioniert, wird aktuell in mehreren deutschen nuklearmedizinischen Praxen und Kliniken getestet [35].…”

Ultraschallgeführte lokalablative Verfahren zur Behandlung von Schilddrüsenläsionen

“…B. mittels tUS, PIUR imaging) Vorteile, da dieser bei der Volumenbestimmung lediglich einen intrinsischen Fehler von ca. 7% aufweist (im Vergleich treten bei der klassischen Drei-Linien-Methode bis zu 40% Fehlerquote auf [37]) und anschaulich für den Patienten mittels zweier 3D-Bilder (vorher/nachher) ausgewertet werden kann (▶ Abb. 7).…”