Key result
High-resolution MRA with deep learning reconstruction links to ~9-point fewer suspected aneurysm classifications.
Why the study?
Conventional MR angiography may yield indeterminate findings when evaluating small suspected intracranial aneurysmal lesions and anatomical variants such as infundibular dilations.
Does high-resolution MR angiography with 0.4 mm isotropic voxels and deep learning-based reconstruction improve lesion classification and diagnostic confidence compared to conventional MR angiography in patients with suspected intracranial aneurysmal lesions?
Observational (n=79)
No
Does high-resolution MR angiography with 0.4 mm isotropic voxels and deep learning-based reconstruction improve lesion classification and diagnostic confidence compared to conventional MR angiography in patients with suspected intracranial aneurysmal lesions?
High-resolution 0.4 mm isotropic MRA with deep learning-based reconstruction improves visualization of microvascular branches, shifting indeterminate intracranial vascular lesions to higher-confidence diagnostic categories and helping differentiate true aneurysms from infundibular dilations.
Answers are grounded in this paper and cite their sources.
May improve diagnostic confidence for suspected intracranial aneurysms; leaves open need for prospective outcome studies before adoption.
PURPOSE: Conventional MR angiography (MRA) may yield indeterminate findings when evaluating small suspected aneurysmal lesions and anatomical variants, such as infundibular dilations (IDs). This study aimed to compare lesion classification and diagnostic confidence between high-resolution MRA with 0.4 mm isotropic voxels reconstructed using deep learning-based reconstruction (DLR) and conventional MRA (C-MRA). METHODS: This study included 79 patients who underwent both C-MRA and high-resolution DLR MRA, with 107 suspected aneurysmal lesions. Two neuroradiologists categorized the lesions into 4 groups: definite aneurysm (dAN), probable aneurysm (pAN), definite ID (dID), and probable ID (pID). RESULTS: On C-MRA, the 107 lesions were classified as 78 dANs, 11 pANs, 5 dIDs, and 13 pIDs. On high-resolution DLR MRA, 5 lesions were reclassified as normal vascular structures. The remaining 102 lesions were reclassified as 78 dANs, 1 pAN, and 23 dIDs. Probable lesions decreased from 24 to 1, whereas definite lesions increased from 83 to 101. McNemar's test revealed a statistically significant shift from the aneurysmal group to the non-aneurysmal group between the 2 methods (P < 0.01). Interobserver agreement (Cohen's kappa) was 0.72 for C-MRA and 0.88 for high-resolution DLR MRA. CONCLUSION: High-resolution MRA with DLR improved visualization of microvascular structures and shifted most indeterminate lesions to higher-confidence categories, suggesting its potential utility in differentiating aneurysms from IDs.
No takes yet. Share an insight, caveat, or question.
Okaneya et al. (2026) conducted an observational in Suspected intracranial aneurysmal lesions (n=79). High-resolution MR angiography with deep learning-based reconstruction (DL-MRA) vs. Conventional MR angiography (C-MRA) was evaluated on Lesion classification into the aneurysmal group (definite or probable aneurysm) among 107 suspected lesions (p=<0.01). High-resolution MRA with deep learning reconstruction reduced low-confidence probable lesions from 24 to 1 and shifted classifications from aneurysmal to non-aneurysmal.
Synapse has enriched 10 related papers. Consider them for comparative context: