Use of artificial neural networks in the analysis of trabecular bone on digitized radiographs

Bone architecture is an important factor that determines bone strength in addition to bone mass. Yet it is only bone mass that is measured in bone mineral densitometry (BMD), which is the most common, clinically used method to assess bone strength. Texture analysis of the trabecular bone pattern on radiographs is being investigated as a potential means to characterize the bone architecture. In this study the authors examined the use of an artificial neural network to merge several texture measures to obtain a single measure related to bone strength. The texture analyses were performed on digitised radiographs of excised femoral necks. Compressive strength measurements of the specimens were used in the training of the ANN. Receiver operating characteristic (ROC) analysis was used to measure the performance of the ANN in distinguishing between strong and weak bone. With direct exposure radiographs, the ANN achieved an area under the ROC curve (A_Z) of 0.98±0.05 in consistency testing and 0.83±0.08 in round-robin analysis. In comparison, BMD measurements on the specimens yielded an A_z value of 0.72±0.11. These results indicate that the texture analysis of trabecular bone pattern on radiographs, merged with the use of an ANN, may be a useful method to noninvasively assess bone strength