Scoliosis is a spinal disorder characterized by a three-dimensional deformity of the vertebral column, often manifesting through observable postural asymmetries. Traditional diagnostic procedures rely heavily on radiographic imaging, measuring the Cobb angle, to specify the spinal curvature severity. However, frequent exposure to radiation is not ideal, especially for pediatric patients. This study proposes a non-invasive, image-based screening method utilizing MediaPipe Pose for the visual detection of postural asymmetries associated with scoliosis, by analyzing back view images of subjects and extracting keypoints to highlight structural irregularities such as elbow, shoulder and wrist height imbalance and misalignment. These observations, validated by a clinical specialist, serve as a preliminary stage for a Bayesian classification approach to distinguish between healthy individuals and those with scoliosis. MediaPipe landmark identification was validated against specialist data, showing no significant differences ( \(p>0.05\) ) on a Kruskal-Wallis test. Further statistical analysis demonstrated highly significant differences when comparing healthy and scoliotic extracted landmarks, confirming their discriminative power. The proposed Bayesian approach achieved accuracies of 0.95 for wrists, 0.9 for elbows, and 0.7 for shoulders. These results underscore the significant potential of this non-invasive, objective method for preliminary scoliosis screening, which also can reduce dependency on radiographs, and serve as a starting point for future simple automated assessment systems.