Early screening and diagnosis of low-grade bladder cancer (LGBC) can help to guide timely clinical treatments before deterioration, reducing relapse rates and improving patient survival and quality of life. However, current clinical technologies are mainly invasive, painful, and lack sensitivity and time efficacy, which cannot always meet clinical needs. Surface-enhanced Raman scattering (SERS) is a label-free detection technique with high sensitivity and can provide molecular-specific information. In this work, we adopt SERSomes, an advanced SERS characterization approach using a SERS spectral set, to comprehensively and accurately profile urine metabolites of LGBC patients and healthy controls. With the help of machine learning, we achieved high accuracy of LGBC diagnosis (89.47%) and LGBC stratification (90%). The entire diagnostic process is very rapid, convenient, non-invasive, and low-cost, holding potential for future use in mass population health screenings. Moreover, we explore the metabolite contribution based on the varying SERSome patterns in LGBC patients, aiming at indicating potential urine biomarkers of LGBC.