Background <p>Carcass composition traits, such as lean meat percentage, bone percentage, and number of ribs, are critical factors determining meat production and profitability of pigs. Traditional slaughter measurements are time-consuming, labor-intensive and invasive and cannot be evaluated on selection candidates. However, computed tomography scanning, a non-invasive technique, enables in vivo measurement of these traits, facilitating rapid accumulation of extensive phenotypic data. Despite these advances, the genetic mechanisms underlying computed tomography-based carcass traits remain largely unexplored.</p> Results <p>In this study, we performed a multi-ancestry genome-wide association meta-analysis (MA-GWAMA) using low-coverage whole-genome sequencing data from four breeds (1222 Duroc, 582 Landrace, 1018 Yorkshire, and 448 Piétrain). In total, we identified 11 independent genome-wide significant loci associated with carcass composition traits in the meta-analysis. Compared to standard genomic best linear unbiased prediction, weighting MA-GWAMA-significant SNPs increased genomic prediction accuracy in an independent population (<i>N</i> = 365, including 136 Duroc, 65 Landrace, 50 Piétrain, and 114 Yorkshire) by 16.3% for lean meat percentage, by 6.1% for bone percentage, and by 79.4% for number of ribs. Integrating MA-GWAMA results with public eQTL and single-cell data prioritized <i>ALPK2</i> as a candidate gene for lean meat percentage, and <i>ABCD4</i> and <i>SLC8A3</i> as candidate genes for the number of ribs.</p> Conclusions <p>Our study demonstrates the efficacy of computed tomography phenotyping coupled with multi-omics integration for dissecting the genetic architecture of porcine carcass composition traits. The prioritized variants and genes provide valuable targets for molecular breeding programs to enhance meat quality in pigs.</p>