This work proposes to learn implicit neural representations of shift-variant PSFs for optical imaging systems that use diffractive optical elements. Results show that the proposed method is able to characterize shift-variant PSFs from sparse samples.