Rapid MR examination with deep learning reconstruction improves diagnosis of complicated acute appendicitis.

OBJECTIVE

To evaluate the performance of a non-contrast rapid magnetic resonance imaging (MRI) protocol with deep learning reconstruction (DLR) for identifying complicated appendicitis in patients with acute appendicitis, compared to conventional MRI without DLR (non-DLR) and non-contrast computed tomography (NC-CT).

METHODS

Sixty-two eligible patients were enrolled, of which 55 had a final diagnosis of acute appendicitis and were included for further analysis of complications. Among them, 51 underwent both NC-CT and MRI, while 4 pregnant patients underwent MRI only. The MRI protocol included T2-weighted imaging (with and without fat suppression), T1-weighted imaging, and diffusion-weighted imaging (DWI). Two abdominal radiologists independently evaluated the subjective and objective image quality of DLR and non-DLR images using a 5-point Likert scale to assess diagnostic confidence for complicated appendicitis. Interobserver agreement (kappa statistics) and receiver operating characteristic (ROC) analysis were performed.

RESULTS

MRI with DLR demonstrated significantly better objective and subjective image quality than non-DLR MRI, with improved appendix delineation and reduced noise (p < 0.01). With a mean acquisition time of 9.5 min, both MRI protocols outperformed NC-CT in the diagnosis of complicated appendicitis (p = 0.035 andp = 0.001, respectively). MRI-DLR achieved higher sensitivity (97.06 % vs. 88.24 %) and negative predictive value (95.45 % vs. 84.00 %) for complicated appendicitis than MRI-non-DLR. The detection rates of appendiceal wall discontinuity and peri-appendiceal abscesses were significantly higher with MRI-DLR (p < 0.05).

CONCLUSION

Rapid MRI with DLR improves image quality and diagnostic accuracy for complicated appendicitis compared to non-DLR MRI and NC-CT, offering a valuable alternative for radiation-sensitive patients.