Multi-task Deep Learning for Cerebrovascular Disease Classification and MRI-to-PET Translation

Accurate cerebral blood flow (CBF) quantification is essential to diagnose and treat many cerebrovascular diseases, including Moyamoya, carotid stenosis, and stroke. The gold standard for CBF (oxygen-15 water positron emission tomography, PET) is not widely available because of its high cost, use of ionizing radiation, and logistical challenges as compared to magnetic resonance imaging (MRI). In this study, using simultaneous PET/MRI, we propose a multi-task learning framework for brain MRI-to-PET translation and disease classification. The proposed framework comprises two prime networks: (1) an attention-based 3D convolutional encoder-decoder network to synthesize high-quality PET CBF maps from multi-contrast MRI images, and (2) a multi-scale 3D convolutional network to identify the brain disease corresponding to the input MRI images. Our multi-task framework yields promising results on the task of MRI-to-PET translation, achieving an average structural similarity index of 0.94 and peak signal-to-noise ratio of 38dB on a cohort of 120 subjects. In addition, we show that integrating multiple MRI modalities can improve the clinical diagnosis of brain diseases. As such, deep learning offers the possibility to perform high-quality CBF measurements and disease classification for patients with cerebrovascular disease at MRI-only sites, leading to improved and more equitable patient care.