Histopathologic correlation of nominal 0.4-T gadoxetic acid-enhanced hepatobiliary-phase MRI with image-domain deep-learning reconstruction in three mice with hepatocellular carcinoma

Technical Summary: Histopathologic Correlation of 0.4-T Gadoxetic Acid-Enhanced MRI with Deep-Learning Reconstruction in Murine HCC

Problem Statement
Gadoxetic acid-enhanced hepatobiliary-phase magnetic resonance imaging (MRI) is a standard modality for characterizing liver lesions, where hepatocellular carcinoma (HCC) typically appears as a hypointense region against the enhanced liver parenchyma. While low-field MRI systems offer increased accessibility, they are historically limited by a reduced signal-to-noise ratio (SNR). Deep-learning reconstruction (DLR) has emerged as a software-based strategy to enhance image quality in lower-field settings. However, for hepatobiliary imaging, the critical challenge is not merely noise reduction but the preservation of lesion conspicuity. There is a lack of descriptive data validating whether nominal 0.4-T gadoxetic acid-enhanced MRI, when combined with image-domain DLR, can produce interpretable images that correlate with histopathology in small animal models of HCC.

Methodology
This study presents a single descriptive visual-pathology observation involving three male mice with primary HCC induced via a neonatal streptozotocin-treated DIAR model.

• Animal Model: Mice were monitored via ultrasound from 8 weeks of age. MRI was performed after hepatic masses were identified or showed enlargement.
• MRI Acquisition: Imaging was conducted on a nominal 0.4-T open MRI system (APERTO Eterna) using a wrist coil. Gadoxetic acid disodium (0.025 mmol/kg) was administered intravenously, with hepatobiliary-phase imaging acquired 20 minutes post-injection. A coronal 3D radiofrequency-spoiled steady-state gradient-echo sequence was used with a 5-minute 33-second acquisition time.
• Deep-Learning Reconstruction (DLR): Standard-reconstructed DICOM images (256 × 256 matrix) were processed using SwiftMR (AIRS Medical Inc.). The reconstruction involved significant post-processing upsampling, increasing the output matrix to 1024 × 1024 with a corresponding reduction in pixel spacing and slice interval (four-fold in-plane and two-fold through-plane upsampling).
• Assessment: A radiologist with 16 years of experience visually compared standard images against DLR-reconstructed images for hypointense hepatic lesions. These findings were correlated with anatomically corresponding hematoxylin and eosin (H&E) stained sections.
• Scope: The study was explicitly designed as a descriptive report. No quantitative metrics (SNR, CNR), diagnostic accuracy calculations, or inter-reader analyses were performed.

Key Results

• Lesion Visibility: In all three mice, hypointense index nodules were visible on both standard and DLR-reconstructed images.
• Histopathologic Correlation: Three index nodules (two slightly larger than 10 mm, one subcentimeter) corresponded anatomically to HCC on H&E staining.
• Effect of DLR: The primary visual difference observed was in the mouse with the subcentimeter lesion (Case 3). On DLR-reconstructed images, an adjacent approximately 1-mm hypointense focus appeared more clearly separated from the index nodule compared to the standard reconstruction. However, histopathology did not confirm this adjacent focus as a separate carcinoma.
• Image Characteristics: The authors note that the apparent sharpness in the DLR images is confounded by the four-fold in-plane and two-fold through-plane upsampling. Consequently, the increased visual definition cannot be equated with increased acquired spatial resolution.

Significance and Claims
The paper claims to support the technical feasibility of obtaining pathology-correlated liver tumor images using nominal 0.4-T gadoxetic acid-enhanced MRI. The study demonstrates that such low-field systems can provide visually interpretable hepatobiliary-phase images that correspond to histopathologic findings of HCC.

However, the authors maintain a modest stance regarding the utility of DLR:

• The study does not establish that DLR improves lesion detection or diagnostic performance, as all three index lesions were already visible on standard reconstructions.
• The only qualitative improvement noted was the clearer visual separation of a small adjacent focus in a single case, which was not histologically confirmed as a separate tumor.
• The authors explicitly caution against interpreting the results as evidence of improved spatial resolution or clinical diagnostic accuracy. They state that the findings should not be extrapolated to human HCC surveillance, minimum detectable lesion sizes, or clinical performance without further controlled studies involving quantitative measures and standard-field MRI comparisons.

In summary, this report serves as a proof-of-concept for the visual-pathologic correlation of low-field MRI in a murine HCC model, while highlighting the interpretative limitations introduced by aggressive post-reconstruction upsampling.