Scoliosis is a common 3D spinal deformity that leads to aesthetic deformity of the torso and can progress to cause pain, osteoarthritis, disability or even respiratory collapse if untreated. The cancer risk associated with the series of spinal X-rays currently used to monitor scoliosis progression would be reduced by use of non-invasive monitoring of torso surface asymmetry. However, the nature of the complex, non-linear relation between surface and spinal deformities has been difficult to quantify. The current project approached the old problem of relating surface asymmetry to scoliotic spinal deformity (measured primarily by the Cobb angle) with a novel combination of three methods:​ 360° full-torso surface scans (rather than the traditional scans of only the back), 3D reconstructions of the spine and rib cage, and non-linear data analysis methods including genetic algorithms and neural networks to interpret the complex relations between these data sets.

Laser scans of 360° torso surface topography were generated and cross-sections cut at closely spaced vertical levels through the resulting surface model. Asymmetry of crosssections was quantified based on the location of specific points on the torso surface as well as geometric properties of each cross-section and of the left and right halves of the section, defined by areas and first and second moments of area. A genetic algorithm was the best of several techniques for selecting the optimal set of torso asymmetry indices and clinical descriptors (e.g., age, weight, and bracing status) used by an artificial neural network to estimate the Cobb angle. The genetic algorithm-neural network combination estimated the Cobb angle within 0.8±5.9° (mean ± standard deviation of difference) in 153 scans of 52 patients, and correctly categorized 92% of patient-records as having mild, moderate or severe curves (Cobb angles <30°, 30-50°, and >50° respectively). That was much more accurate than previous attempts to relate surface and spinal deformity and justified future longitudinal studies to estimate scoliosis progression from changes in torso asymmetry in individual patients using similar data reduction techniques.