Hybridization involving extinct or unsampled ("ghost") lineages profoundly influences species' evolutionary histories, but detecting such introgression remains methodologically challenging. We introduce D-BPP, a novel framework that integrates the heuristic D-statistic (or ABBA-BABA test) with Bayesian phylogenomic inference (implemented in BPP) to efficiently infer phylogenetic networks. In D-BPP, we first employ the D-statistic to rapidly identify candidate introgression events on a predefined binary species tree; then we leverage the Bayesian test in BPP to rigorously validate these candidates and sequentially add them to the species tree, retaining only those events with strong statistical support. If the species tree is ambiguous, D-BPP identifies the most probable tree by evaluating competing topologies through Bayesian model comparison of their corresponding introgression models. Critically, our framework excels at detecting ghost introgression, which is often unidentifiable or overlooked by existing methods-whether heuristic or full-likelihood. Applied to genomic datasets from Panthera (big cats) and Thuja (conifers), D-BPP uncovered previously undetected ghost introgression events in both clades, underscoring the pervasive role ghost lineages have played across diverse taxa. By uniquely combining the computational efficiency of heuristic D-statistics with the robust statistical rigor of full-likelihood Bayesian inference, D-BPP deciphers complex hybridization patterns obscured by conventional methods, providing a powerful tool for accurately reconstructing phylogenetic networks.