Early and accurate prediction of lymph node metastasis (LNM) in endometrial cancer (EC) is critical for improving patient outcomes. A predictive molecular signature for LNM could enable preoperative staging and risk assessment, reducing the need for unnecessary surgical procedures and minimizing gynaecological cancer morbidity. In this study, we employ statistical machine learning techniques that integrate variable ranking with canonical correlation (CC) analysis into a classification framework. Using tissue microarrays (TMA) and MALDI mass spectrometry imaging (MSI) data from 172 patients, our CC-based discrimination approach achieves 98.3% accuracy at the individual spectra level and 100% accuracy at the patient level in classifying LNM status. Furthermore, we extend our approach to curettage tissue samples from 34 patients, which were collected non-invasively. When using the 40 highest-ranked masses, we achieve the same 100% classification accuracy at the patient level. These findings highlight the potential of non-invasive molecular diagnostics for early metastasis detection, reducing the need for invasive surgical staging and enabling more targeted and personalized treatment strategies for endometrial cancer patients.