Acute Leukemia Diagnosis Through AI-Enhanced Attenuated Total Reflection Fourier Transform Infrared Spectroscopy of...

Acute leukemia, a highly perilous cancer, is diagnosed using invasive procedures like bone marrow aspirate and biopsy (BMA/BMB). This study investigated the use of artificial intelligence (AI)-enhanced Fourier transform infrared (FT-IR) spectroscopy as a non-invasive, reagent-free diagnostic alternative with high sensitivity and specificity. The spectral peak patterns of peripheral blood smears (PBS) from clinically healthy individuals (n = 50) BMA/BMB-confirmed acute leukemia patients (n = 50) were examined in the 1800-850 cm^-1 range. Six trained models were used to assess the diagnostic performance, focusing on accuracy, positive predictive value (PPV), negative predictive value (NPV), F₁ score, and area under the receiver operating characteristic (ROC) curve (AUC). The study shows significantly lower absorbance peaks in leukemia cases compared to healthy controls across various spectral regions: 1637.82, 1528.63, 1448.29, and 1388.54 cm^-1, 1302.02, and 1240.21 cm^-1, and 1163.99 cm^-1. These differences indicate decreased concentrations or distinct molecular configurations of proteins, lipids, nucleic acids, and carbohydrates in cases. Conversely, they exhibited elevated absorbance peaks at 1032.14 and 894.11 cm^-1 regions, suggesting potential disparities in amino acid, DNA, fatty acid, and saccharide residues compared to healthy controls. Of the six trained models, the SVM model demonstrated remarkable diagnostic performance, achieving an accuracy of 83%, a PPV of 80%, an NPV of 86%, an F₁ score of 82.47%, and an AUC of 90.76%. This study demonstrates the potential of AI-enhanced FT-IR spectroscopy as a valuable adjunct diagnostic tool for acute leukemia. By offering a less invasive and faster alternative to BMA/BMB, this approach can potentially enhance leukemia diagnosis and improve patient outcomes, particularly in pediatric and geriatric cases.