In the context of viral epidemic surveillance, generating accurate consensus viral genomes from sequencing data is critical for tracking the emergence of mutations of concern, evaluating the genomic diversity of circulating viruses, and anticipating which viral strains could become most prevalent. However, this task is made difficult by the presence of Deletion-containing Viral Genomes (DelVGs), which contain truncated (or rearranged) and potentially mutated versions of the full length virus genome. Because these DelVGs can outnumber the full genome in terms of coverage, potential DelVG specific mutations may be erroneously incorporated into the final consensus, thereby compromising its accuracy. Automatic detection of these DelVGs and of the genomic positions that may harbor DelVG specific mutations is therefore crucial. Here, we present DIPScan, a new method able to (i) accurately and efficiently detect DelVGs in short read datasets, and (ii) mask or correct positions in the consensus genome that may be affected by DelVG-specific mutations. DIPScan achieves this through tailored metrics for breakpoint characterization and selection, linear modeling to estimate DelVG relative abundance from well-defined region and junction coverage, and efficient heuristic algorithms for reliable consensus sequence correction. Using several hundreds of simulated and real patient-derived NGS datasets from the National Reference Center (NRC) for respiratory viruses at Institut Pasteur, we demonstrate the capacity of DIPScan to accurately and efficiently detect DelVGs and to correctly adjust the consensus sequences. DIPScan is implemented as a Nextflow workflow, making it highly flexible, scalable, and reproducible, and is now used routinely at the NRC.