Exercise‐induced calf muscle hyperemia: Rapid mapping of magnetic resonance imaging using deep learning approach

Abstract

Exercise‐induced hyperemia in calf muscles was recently shown to be quantifiable with high‐resolution magnetic resonance imaging (MRI). However, processing of the MRI data to obtain muscle‐perfusion maps is time‐consuming. This study proposes to substantially accelerate the mapping of muscle perfusion using a deep‐learning method called artificial neural network (NN). Forty‐eight MRI scans were acquired from 21 healthy subjects and patients with peripheral artery disease (PAD). For optimal training of NN, different training‐data sets were compared, investigating the effect of data diversity and reference perfusion accuracy. Reference perfusion was estimated by tracer kinetic model fitting initialized with multiple values (multigrid model fitting). Result: The NN method was much faster than tracer kinetic model fitting. To generate a perfusion map of matrix 128 × 128 on a same computer, multigrid model fitting took about 80 min, single‐grid or regular model fitting about 3 min, while the NN method took about 1 s. Compared to the reference values, NN trained with a diverse group gave estimates with mean absolute error (MAE) of 15.9 ml/min/100g and correlation coefficient (R) of 0.949, significantly more accurate than regular model fitting (MAE 22.3 ml/min/100g, R 0.889, p < .001). Conclusion: the NN method enables rapid perfusion mapping, and if properly trained, estimates perfusion with accuracy comparable to multigrid model fitting.

Machine learning in vascular surgery: a systematic review and critical appraisal

“…Twenty performed better, two performed similarly, and none performed worse. Specifically, 6 performed better than traditional regression models such as logistic, linear, and Cox regression 16,[27][28][29][30][31] , 11 performed better than existing risk prediction tools such as the Glasgow Aneurysm Score, Mangled Extremity Severity Score (MESS), and Padua Prediction Score [32][33][34][35][36][37][38][39][40][41][42] , 1 performed better than vascular surgeons in predicting inhospital mortality following AAA repair 43 , and 2 performed better than radiologists in detecting AAA on CT 15,44 . One performed similarly to logistic regression for predicting shunt necessity during carotid endarterectomy 45 and another demonstrated no difference compared to radiologists in detecting aortic dissection on CT 46 .…”

Leveraging Machine Learning and Artificial Intelligence to Improve Peripheral Artery Disease Detection, Treatment, and Outcomes

“…B and C , Computer vision algorithms may also enable more efficient processing of magnetic resonance imaging (MRI) images without compromising accuracy. Computer vision models generate accurate calf muscle perfusion maps in <1 s, compared with 180 min by standard modeling ( B , modified from Zhang et al 73 with permission. Copyright ©2020, Wiley).…”

“…Innovative studies in MRI analysis have also been recently reported and may reduce the time and labor-intensive processing of MRI images. Zhang et al 73 developed a computer vision model to accelerate mapping of calf muscle perfusion from dynamic contrast-enhanced MRIs. Their feedforward neural network was developed using 48 MRI scans, including pre- and post-exercise scans from healthy subjects and patients with PAD.…”

Personalized Cell Therapy for Patients with Peripheral Arterial Diseases in the Context of Genetic Alterations: Artificial Intelligence-Based Responder and Non-Responder Prediction

“…As previously mentioned, the analysis of medical images is one of the greatest success stories of AI, spanning the analysis of histopathological images [ 113 ], electrocardiograms [ 114 ], radiographs [ 115 ], magnetic resonance imaging slices [ 116 ], and many more [ 108 ]. Since imaging is very much an emerging field in PAD, this bears high potential because several already-established biomarkers could already be used for an unbiased patient stratification [ 117 ].…”