Serum Fourier-Transform Infrared Spectroscopy with Machine Learning for Screening of Pediatric Acute Lymphoblastic Leukemia: A Proof-of-Concept Study

Background: Acute lymphoblastic leukemia (ALL) is the most common childhood malignancy, yet diagnosis still relies primarily on invasive bone-marrow procedures and advanced laboratory assays. Non-invasive, rapid, and cost-effective tools remain an unmet need. Fourier-transform infrared (FTIR) spectroscopy has shown promise for detecting cancer-associated biochemical changes in biofluids and cells. Methods: Serum from pediatric ALL patients and controls (n = 103; ALL = 45, controls = 58: healthy = 14, hematology controls = 44 with anemia, thrombocytopenia, leukopenia, and pancytopenia) was analyzed using FTIR. Spectra (800–1800, 2800–3500 cm−1) were preprocessed with baseline correction, derivative filtering, and normalization. Group differences were assessed statistically, and logistic regression with stratified 10-fold cross-validation was applied; Receiver operating characteristic (ROC)\precision–recall (PR) analyses were based on out-of-fold predictions. Results: Distinct spectral alterations were observed between ALL and controls. Leukemia samples showed higher amide I (~1640 cm−1) and amide II (~1545 cm−1) absorbance, lower lipid-related bands (~1450, ~2920 cm−1), and increased nucleic-acid–associated signals (~1080 cm−1). Differences were significant (q < 0.05) with moderate effect sizes. Logistic regression achieved area under the curve (AUC) ≈ 0.80 with sensitivity ~0.73–0.84 across practical decision thresholds (0.50 → 0.30) and higher recall attainable at the expense of specificity. Principal component analysis (PCA)\hierarchical cluster analysis (HCA) indicated partial but consistent group separation, aligning with supervised performance. Conclusions: Serum FTIR spectroscopy shows promise for distinguishing pediatric ALL from controls by reflecting disease-related metabolic changes. The technique is rapid, label-free, and requires only small serum volumes. Our findings represent proof-of-concept, and validation in larger, multi-center studies is needed before clinical implementation can be considered.