Advances in magnetic resonance imaging of peripheral arterial disease: non-contrast magnetic resonance angiography and artificial intelligence-aided plaque analysis
Csőre Judit
Cardiovascular Medicine and Research Division
Dr. Merkely Béla
SE Városmajori Szív- és Érgyógyászati Klinika
2024-07-02 13:30:00
Klinikai és kísérlettes angiológiai kutatások
Csobay-Novák Csaba
Dr. Berczeli Márton
Dr. Trajtler Andrea
Dr. Vágó Hajnalka
Dr. Korda Dávid
Dr. Piróth Zsolt
Dr. Lakatos Bálint
Introduction: Peripheral arterial disease (PAD) is often associated with impaired renal function, underscoring the necessity for nephroprotective modalities in assessment. Our study compared iodinated contrast digital subtraction angiography (DSA) and carbon dioxide (CO2) DSA with non-contrast Quiescent-Interval Single-Shot Magnetic Resonance Angiography (QISS MRA). We also assessed the feasibility of a custom variational autoencoder (VAE) using 2D convolutional neural networks (CNNs) on 7T magnetic resonance (MRI) images to differentiate soft vs. hard plaque components in PAD.
Methods: Patients underwent elective iodinated or CO2 contrast DSA and 1.5T QISS MRA. Stenoses and image quality were evaluated. Inter-rater reliability was calculated using intraclass correlation coefficient (ICC). Diagnostic accuracy and interpretability of the modalities were assessed. VAE was evaluated on 7T MRI scans of amputated lower extremities utilizing tissue scores and classes (Tissue Class I: lumen patent, tissue score 0; Tissue Class II: lumen partially patent, tissue score 1; Tissue Class III: lumen mostly occluded with soft tissue, tissue score 3 and Tissue Class IV: lumen mostly occluded with hard tissue, tissue score 5)
Results: We analyzed 34 patients (623 segments) for iodinated DSA vs. QISS MRA and 28 patients (523 segments) for CO2 DSA vs. QISS MRA. QISS MRA demonstrated superior image quality over CO2 DSA in all regions (p<0.001) and was comparable in image quality to iodinated DSA while surpassing both in interpretability (p<0.001). Compared to iodinated DSA, diagnostic accuracy parameters for QISS MRA were: diagnostic accuracy 91.3%, sensitivity 84.8%, specificity 93.0%, PPV 76.3%, NPV 95.8%. Compared to CO2 DSA, QISS MRA had a diagnostic accuracy of 91.1%, sensitivity 77.8%, specificity 95.2%, PPV 83.2%, NPV 93.3%. Interobserver variability for stenosis assessment in all regions combined was 0.94 for QISS MRA and 0.88 for iodinated DSA; for QISS MRA and CO2 DSA, it was 0.97 and 0.82, respectively (p<0.001). We utilized 2390 MPR reconstructed MRI images to train our algorithm. Relative percentage of average tissue score varied from completely patent (lesion #1) to the presence of all four tissue classes. Lesions #2, #3 and #5 were classified to contain all tissue classes except Tissue Class IV, while lesion #4 contained all classes. Training the VAE was successful, as images with soft/hard tissues in PAD lesions were satisfactorily separated in latent space.
Conclusion: QISS MRA is a reliable and safe modality for PAD imaging. Also, using VAE may assist in rapid classification of MRI histology images acquired in a clinical setup for facilitating therapeutic decision-making.
