Yang et al. conducted a prospective, multicenter clinical study involving 301 patients with gallbladder-occupying lesions (GBOLs) to develop a non-invasive model capable of distinguishing gallbladder cancer (GBC) from benign lesions (GBBLs). Using blood-based targeted next-generation sequencing of 111 genes and CT imaging, the researchers built three diagnostic models: a cfDNA-based model (CBM), a radiomic model (RM), and an integrated multimodal classifier called GBCseeker. In the discovery cohort, cfDNA from GBC patients exhibited significantly higher concentrations, mutation frequency, and variant allelic fractions than GBBLs. TP53 mutations were exclusive to GBC. The CBM, incorporating 14 genomic variables along with CA19-9 levels and sex, achieved an AUC of 0.94 in the training set and 0.86 in the test set; however, it demonstrated moderate specificity (73.7%). The RM, built from 42,484 CT slices and automated segmentation, showed better specificity (84.2%) and an AUC of 0.94. GBCseeker, which combines CBM and RM scores through logistic regression, achieved the highest performance with an AUC of 0.97 in the test set and 0.93 in an external validation cohort of 98 patients. GBCseeker significantly outperformed surgeons in identifying early-stage (AJCC 0–I) GBC (sensitivity increased from 57.8% to 78.9%) and reduced overall diagnostic error by 56.2%. Furthermore, the model reclassified patients into low-, medium-, and high-risk categories to guide surgical management, correctly classifying 91.8% of GBBLs and preventing overtreatment in 88.6% of cases.