Radio interferometry has historically relied on invariants like closure phases and closure amplitudes, which are immune to antenna-based calibration and errors therein, and therefore contain true morphological information about the objects of interest. However, a direct correspondence between the closure invariants and morphological features towards image reconstruction has not been established to date. Beginning with a review of the theory of closure invariants, I will describe recent work using machine learning that successfully estimates parameters of various morphological classes, and subsequently reconstructs images using closure invariants alone without employing any prior assumptions. The technique promises to provide not only an insight into an object's morphological features, but also provide an independent and complementary path to probing event horizon scale morphology around black holes.