The nature of deep earthquakes with depths greater than 70 km is enigmatic because brittle failure at this high-temperature and the high-pressure regime should be inhibited. Three main hypotheses have been proposed to explain what causes deep earthquakes within the subducting slabs, dehydration embrittlement, phase transformational faulting, and thermal runaway instability. However, the existing seismological constraints can’t yet definitively distinguish between these hypotheses because the fine 3D slab structures are not well constrained in terms of slab upper interface, thickness, and internal fine layering. To better image the slabs in the Western Pacific subduction zones, this study employs a full waveform inversion (FWI) that minimizes waveform shape misfit between the synthetics and the observed waveforms from a large dataset, with 142 earthquakes recorded by about 2,400 broadband stations in East Asia. A 3-D initial model that combines two previous FWI models in East Asia (i.e., FWEA18 and EARA2014) are iteratively updated by minimizing the misfit measured from both body waves (8–40 s) and surface waves (30–120 s). Compared to the previous models, the new FWI model (EARA2020) shows much stronger wave speed perturbations within the imaged slabs with respect to the ambient mantle, with maximum perturbation of 8% for Vp and 13% for Vs. Furthermore, the slab thickness derived from EARA2020 exhibits significant downdip and along-strike variations at depths greater than 100 km. The large intra-slab deep earthquakes (Mw>6.0) appear to occur where significant slab thinning happens. This observation suggests that the significant deformation (or strain accumulation) of the slab is likely the first-order factor that controls the distribution of large deep earthquakes within the slab regardless of their triggering mechanism.