Introduction

Colorectal cancer is the third most prevalent cancer worldwide, with rectal cancer constituting approximately one-third of these cases ¹. Alarmingly, this proportion continues to rise ², exacerbating the disease burden of rectal cancer globally. Magnetic Resonance Imaging (MRI) serves as the preferred and standard imaging modality for rectal cancer assessment. Acquiring superior image quality in shorter time, without compromising data integrity has become an imminent issue in need of resolution to satisfy the growing disease burden of rectal cancer. Deep learning reconstruction (DLR) has demonstrated its ability to enhance the image quality of accelerated MRI in multiple organs ^3,4. However, its application to rectal MRI remained rarely reported. The rectum, characterized as a hollow organ, features a unique anatomical structure, which makes its image susceptible to the functional peristalsis and dynamics contraction and relaxation of the intestinal wall. Therefore, assessing the applicability of DLR for rectal MRI in real clinical scenarios is imperative.
This study aimed to provide comprehensive perspective on the practical application of DLR-based rectal MRI in comparison to standard MRI, encompassing image acquisition time, image quality, in particular, diagnostic performance and reding efficiency.

Methods

Patients: With approval from the institutional review board, consecutive patients with biopsy-proven primary rectal adenocarcinoma between November/2022 and May/2023 were prospectively included. Imaging: Two key sets of oblique axial T2WI using standard FSE (FSE_standard) and DLR-based accelerated FSE (FSE_DL) performed using a 3.0-T MRI system (AIR™ Architect; GE Healthcare, Milwaukee, Wis) were obtained for comparative analysis. Evaluation methods: Imaging quality assessment was conducted using quantitative and qualitative methods and evaluated separately for two protocols. Quantitative image quality assessment was based on region-of-interest measurements. Three parameters including signal-noise ratio (SNR) of the tumor (SNR_tumor), SNR of the internal obturator muscle (SNR_{internal obturator muscle}), and contrast-noise ratio (CNR) between the tumor and internal obturator muscle were calculated respectively for two sequences. Qualitative image quality metrics regarding artifacts, noise, tumor margin clarity, visualization of bowel wall layering, visualization of mesorectal fascia, overall image quality, and diagnostic confidence were assessed based on a five-point Likert scale, whereby larger scores indicate better: 1 – non-diagnostic; 2 – poor but still interpretable; 3 – acceptable; 4 – good; 5 – excellent. Patients undergoing curative surgery without preoperative treatment were re-selected for diagnostic performance and reading efficiency evaluation. Diagnostic assessment included T-staging, N-staging, extramural vascular invasion (EMVI), and mesorectal fascia (MRF) involvement, which was conducted by five radiologists with different levels of experience, as compared with surgical pathology results. To mitigate subjective recall bias, the FSE_standard and FSE_DL imaging was analyzed separately in a random and mixed order, with at least three weeks of washout period between two reviews. Each consuming time of four diagnostic metrics (mrT staging, mrN staging, mrEMVI, and mrMRF) and total time summing up the four metrics in two sequences (FSE_standard and FSE_DL) was recorded for five readers for reading efficiency analysis. Data processing: Comparison of the imaging quality and diagnostic time between FSE_standard and FSE_DL was conducted using paired-sample t-test or Wilcoxon signed-rank test. The diagnostic efficiency was compared by McNemar's test. Intra- and inter-reader agreement was assessed by intraclass correlation coefficient, Cohen’ s κ and Fleiss’ κ.

Results

Totally, 117 patients (77 males; age range 21 – 77 years) were enrolled, with 60 patients (36 males; age range 32 – 77 years) undergoing radical surgery. DLR reduced acquisition time by 65% than standard MRI. Moderate to excellent intra- and inter-reader agreement was achieved for all assessment metrics. FSE_DL exhibited higher SNR_tumor, SNR_{internal obturator muscle}, CNR, and subjective scores related to noise, tumor margin clarity, visualization of bowel wall layering and mesorectal fascia, overall image quality, and diagnostic confidence than FSE_standard (p < 0.001). Reduced artifacts were observed in FSE_DL compared to FSE_standard for patients without spasmolytics (p < 0.05). FSE_DL was rated higher accuracy of T-staging than FSE_standard by junior readers (reader 1, 58% vs. 70%, p = 0.016; reader 3, 60% vs.76%, p = 0.021), with comparable performance in evaluating N-staging, EMVI, and MRF. No significant difference was found concerning all the diagnostic performance by senior readers (p > 0.05). FSE_DL exhibited shorter diagnostic time in T-staging and overall evaluation by all readers, as well as in EMVI and MRF by junior readers (p < 0.05).

Discussion and Conclusions

FSE_DL was clinically viable for rectal MRI, which could facilitate substantial improvements in image quality and reading efficiency than FSE_standard, with a 65% reduction in acquisition time. Moreover, it had the potential to assist junior radiologists in enhancing the diagnostic accuracy of T-staging.