Background: The cfDNA targeted capture sequencing technology has been gradually applied to research central nervous system (CNS) metastases. But targeted sequencing usually requires at least 5 ng of cfDNA. Due to the limited amount of cerebrospinal fluid (CSF) that can be collected there are not many samples that can meet the requirements of targeted sequencing. Methods: This study assessed the use of shallow whole-genomic sequencing (sWGS) of blood and CSF samples obtained from 35 patients with lung cancer to identify tumor-derived cfDNA. Among them, 6 cases had no intracranial metastasis, 22 had only PMs, and 7 had LMs. Some of these cases underwent irregular follow-up. For blood samples, 5 ng cfDNA was used for library preparation. For CSF samples, try to use trace DNA to prepare libraries, regardless of whether the Qubit assays can quantify the concentration. Cytologic analysis was also performed on CSF samples from the LM cases. A total of 35 peripheral blood samples and 31 cerebrospinal fluid samples were subjected to sWGS tests, and the arm-level CNV and tumor proportion were analyzed.Cytologic analysis was also performed on CSF samples from the LMs. Results: The lowest concentration of cfDNA in 35 plasma samples was 2.20 ng/mL. But when the cfDNA was extracted from the CSF, 50% (6/12) of the LM samples and 70.59% (12/17) of the samples with only PMs could not be quantified by Qubit assays. All libraries were all successfully prepared from the samples that could not quantify cfDNA. For CSF samples from LM cases, the sensitivity of detecting tumor signals by the sWGS test was 90.9% (10/11), higher than 81.8% (9/11) by CSF cytologic analysis. For samples from PM cases, the sensitivity of detecting tumor signals by the sWGS test was 75% (21/28); among them, the detection sensitivity of samples with a total amount of cfDNA not less than 5 ng was 90% (9/10), while the detection sensitivity of samples with the total amount of cfDNA less than 5 ng was 66.67% (12/18). Three patients underwent multiple CSF cfDNA tests, of which one case with RANO evaluation as PR had a decrease in the tumor fraction in CSF cfDNA, while only one of the two cases with RANO SD evaluation had stable tumor fraction in CSF cfDNA. Conclusions: We established a method for detecting CNVs by sWGS using trace amounts of cfDNA. Using the sWGS test improves the sensitivity of the detection of LMs. Compared with targeted capture sequencing, our method reduces the requirement for the amount of cfDNA, which can be more widely used in the detection of CNS metastases, especially in PM-only patients with extremely low CSF cfDNA concentrations.