Musculoskeletal (MSK) diseases are widespread, with the World Health Organization estimating in 2019 that 1.71 billion people worldwide are afflicted with the condition [1]. MSK conditions include low back pain, knee osteoarthritis, and rheumatoid arthritis, among others, all of which induce debilitating pain and require early diagnosis to improve prognosis of treatment outcomes. Imaging is a crucial tool for diagnosis, and among available options, Magnetic Resonance Imaging (MRI) is a preferred modality for its sharp soft-tissue contrast, high-resolution images, and lack of ionizing radiation. However, acquisition and processing of MR images has numerous challenges: (1) acquisitions are time-consuming, and therefore expensive and susceptible to motion artifacts; (2) special sequences require toxic contrast agent administration, which have safety concerns; and (3) analysis of acquired images to identify patients most requiring clinical intervention is laborious. This work proposes using deep learning to address various aspects of these challenges. I will be presenting 5 applications and uses of deep learning algorithms:1. To accelerate a 3D fat-suppressed knee MR sequence, showing that optimizing reconstruction algorithms for one tissue of clinical interest can improve its performance in other tissues of clinical interest. 2. For image reconstruction of accelerated compositional MR acquisitions in the knee, hip and lumbar spine, optimizing reconstructed images for tissues of heightened clinical interest (cartilage and intervertebral discs). 3. To automatically segment bone and cartilage from 8X accelerated knee MR acquisitions. 4. To synthesize post-contrast wrist MR images from pre-contrast scans in rheumatoid arthritis patients. 5. To predict if patients would require a total knee replacement within 5 years, using MR imaging and demographic variables.