Deep learning reconstruction improves appendix visualization on pediatric magnetic resonance imaging (MRI): a single-center experience.
Authors
Affiliations (8)
Affiliations (8)
- Massachusetts General Hospital/Mass General Brigham, Boston, MA, USA. [email protected].
- Harvard Medical School, Boston, MA, USA. [email protected].
- Massachusetts General Hospital/Mass General Brigham, Boston, MA, USA.
- UMass Memorial Medical Center, Worcester, MA, USA.
- Harvard Medical School, Boston, MA, USA.
- General Electric Healthcare, Chicago, IL, USA.
- Massachusetts General Hospital/Mass General Brigham, Boston, MA, USA. [email protected].
- Harvard Medical School, Boston, MA, USA. [email protected].
Abstract
Thin-slice MRI may improve appendix visualization in children with suspected appendicitis, but reducing slice thickness decreases signal-to-noise ratio (SNR) and can degrade image quality. Deep learning reconstruction (DLR) enables denoising of thin-slice images and may mitigate these limitations. To evaluate whether 2-mm T2-weighted magnetic resonance imaging (MRI) with DLR improves appendix visualization and diagnostic performance compared with conventional 4-mm MRI in pediatric appendicitis. In this prospective, Institutional Review Board-approved, single-center study (January 2022-June 2025), 82 pediatric patients (mean age 9.6 years; age range 5-20 years) undergoing MRI for suspected appendicitis were included. MRI protocols contained both standard 4-mm and thin-slice 2-mm single-shot T2-weighted images; DLR was applied to 2-mm images (2-mm+DLR). Three blinded radiologists independently assessed paired series for appendix visualization, confidence, appendicitis diagnosis, ancillary findings, and overall image quality. Objective SNR was measured in psoas muscle, bladder, and fat. Diagnostic performance was calculated against surgical or clinical reference standards using paired statistical tests. In the first 16 exploratory cases, quantitative analysis showed that the signal-to-noise ratio of 2-mm+DLR images was comparable to that of standard 4-mm images. Across the full cohort, pooled analysis showed improved appendix visualization with 2-mm+DLR compared with standard 4-mm MRI (88.2% [217/246] vs 83.3% [205/246], P=0.006). Overall diagnostic image quality was also higher with 2-mm+DLR (median 5 vs 4, P<0.001), while confidence in appendix visualization and appendicitis diagnosis was similar between techniques (both medians 4, P>0.1). Exploratory pooled diagnostic performance numerically favored 2-mm+DLR, with higher sensitivity (89% [78/87] vs 81% [71/87]) and accuracy (89% [218/246] vs 86% [210/246]), although paired comparisons did not reach statistical significance (Table 4). DLR-enhanced 2-mm MRI improves appendix visualization and overall diagnostic image quality in pediatric appendicitis without reducing reader confidence or specificity. Thin-slice DLR MRI supports improved depiction of the appendix and alternative abdominal pathology in children.