1013. Comparison Deep Learning-Based Reconstructed Dynamic Contrast-Enhanced Pituitary MRI with Conventional Dynamic Contrast-Enhanced Images
Authors * Denotes Presenting Author
  1. Inseon Ryoo; Korea University Guro Hospital
  2. Hye Na Jung *; Korea University Guro Hospital
  3. Sang-il Suh; Korea University Guro Hospital
  4. Leehi Joo; Korea University Guro Hospital
To compare diagnostic performance and image quality between a deep learning-based reconstructed dynamic contrast-enhanced (DLR-DCE) (slice thickness, 2-mm with no gap; temporal sampling interval, 27.5s) and a conventional dynamic contrast-enhanced (C-DCE) (slice thickness, 2-mm with 0.4 mm gap; temporal sampling interval, 40s) images in the evaluation of sellar and suprasellar lesion.

Materials and Methods:
For this retrospective study, 43 consecutive patients underwent combined imaging protocol including 6 dynamic phases of DLR-DCE and following single phase of conventional DCE images between April and August 2022. Two board certified neuroradiologists independently diagnosed the lesions (pituitary adenoma; Rathke’s cleft cyst; postoperative change; craniopharyngioma; no lesion) and identified cavernous sinus invasion using last phase imaging set of DLR-DCE and single phase imaging set of C-DCE. Also they evaluated the image quality between the two imaging sets. Other two board certified neuroradiologists established for correct diagnosis for reference standards by using all available imaging resources, clinical histories, surgical records, and pathology reports.

The diagnostic performance of DLR-DCE was comparable that of C-DCE in the diagnosis of lesion {accuarate diagnosis rate, 81.4% (35/43) vs 72,1% (31/ 43), respectively} and identifying cavernous sinus invasion (area under the receiver operating characteristic curve, 0.84–0.96 vs 0.71–0.94, respectively; P=0.09, respectively). Interreader agreement was ranges from good to excellent (intraclass correlation coefficient, 0.702–0.924). The DLR-DCE outperformed the C-DCE in terms of margin of the normal gland, pulsation artifact, and sharpness (P < 0.03).

In routine practical setting, DLR-DCE demonstrated comparable diagnostic performance in the diagnosis of the sellar and suprasellar lesions, and identification of cavernous sinus invasion to C-DCE. The DLR-DCE provided better imaging quality despite the shorter temporal sampling interval than C-DCE.