Comparative study of different artificial intelligence (AI)-assisted compressed sensing factors in inner ear heavily...

Artificial intelligence (AI)-assisted compressed sensing (ACS) is a cutting-edge magnetic resonance imaging (MRI) acceleration technique based on deep learning. High-quality imaging of the inner ear suffers from lengthy acquisition time due to intricate and fine structure. The study aimed to compare the image quality of heavily T2-weighted inner ear imaging among ACS groups with different accelerated factors (AFs).

A total of 40 healthy volunteers were prospectively included. Using a heavily T2-weighted three-dimensional fast spin-echo sequence, images were acquired in a reference group (without acceleration) and four ACS groups, with AFs of 2.7, 3.87, 4.68, and 5.52, respectively. Image quality was qualitatively assessed using a four-point scale. The signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and edge sharpness were compared across groups.

Compared with the reference group (acquisition time: 354 secs), the acquisition times for the ACS_2.7, ACS_3.87, ACS_4.68, and ACS_5.52 groups were shortened to 39.83%, 28.81%, 23.73%, and 20.62%, respectively. The ACS3.87 group demonstrated superior imaging quality compared with the reference group and the ACS4.68 and ACS5.52 groups, but showed no significant difference from the ACS2.7 groups. The ACS3.87 group outperformed all other groups with respect to SNR and CNR. Regarding edge sharpness, it was only significantly better than that of the reference group.

The ACS_3.87 group improved imaging quality while substantially reduced acquisition time compared with other ACS groups. An AF of approximately 3.5- to 4-fold is recommended for clinical application in heavily T2-weighted inner ear imaging.