Fig. 1.

Crystal Bone: Use of Machine Learning driven methods for fracture prediction. Crystal Bone used techniques applied in natural language processing to screen electronic healthcare records from a US population (Optum), covering 91 million patients to predict first and second fracture at the spine, pelvis, clavicle, humerus, radius, ulna, hip, femur, tibia, fibula, and ankle. Sequences of ICD codes were used as inputs to implement two distinct frameworks: (1) ICD code vectorization and long short-term memory networks, and (2) patient-level vectorization and extreme gradient boosting decision trees. The figure shows exploration of model interpretability by comparison of various characteristics of the input data for the 4 prediction cohorts of the confusion matrix (FN false negative, FP false positive, TN true negative, TP true positive). UMAP: uniform manifold approximation and projection (this allows encoded vectors to be projected onto a 2D space for dimension reduction). ICD codes predictive of future fracture (TP) include, for example 73,313 (collapsed vertebra), 81,200 (closed fracture of upper humerus), 81,342 (closed fracture of distal radius), 82,100 (closed fracture of femur). Reproduced with permission from [29]. https://www.jmir.org/2020/10/e22550/