The Human T-cell leukemia virus type-I (HTLV-1) is the causal agent of HTLV-associated myelopathy/Tropical Spastic Paraparesis (HAM/TSP). HAM/TSP is the result of demyelination and cell death in the spinal cord and disruption of the blood-brain barrier (BBB), mediated by a virus-induced inflammatory response. In this study, we applied Positron Emission Tomography with 18F-fluordeoxyglucose (18F-FDG PET) to evaluate brain metabolism in a group of 47 patients infected with HTLV-1, and 18 healthy controls. Patients were divided into three groups according to their neurological symptoms. A machine learning (ML) based Gaussian Processes classification algorithm (GPC) was applied to classify between patient groups and controls and also to organize the three patient groups, based on gray and white matter brain metabolism. We found that GPC was able to differentiate the HAM/TSP group from controls with 85% accuracy (p = 0.003) and the asymptomatic seropositive patients from controls with 85.7% accuracy (p = 0.001). The weight map suggests diffuse cortical hypometabolism in both patient groups when compared to controls. We also found that the GPC could separate the asymptomatic HTLV-1 patients from the HAM/TSP patients, but with a lower accuracy (72.7%, p = 0.026). The weight map suggests a diffuse pattern of lower metabolism in the asymptomatic group when compared to the HAM/TSP group. These results are compatible with distinctive patterns of glucose uptake into the brain of HTLV-1 patients, including those without neurological symptoms, which differentiate them from controls. Furthermore, our results might unveil surprising aspects of the pathophysiology of HAM/TSP and related diseases, as well as new therapeutic strategies.