Sparse Annotations with Random Walks for U-Net Segmentation of Biodegradable Bone Implants in Synchrotron Microtomograms

Abstract

Currently, most bone implants used in orthopedics and traumatology are non-degradable and may need to be surgically removed later on e.g. in the case of children. This removal is associated with health risks which could be minimized by using biodegradable implants. Therefore, research on magnesium-based implants is ongoing, which can be objectively quantified through synchrotron radiation microtomography and subsequent image analysis. In order to evaluate the suitability of these materials, e.g. their stability over time, accurate pixelwise segmentations of these high-resolution scans are necessary. The fully-convolutional U-Net architecture achieves a Dice coefficient of 0.750 ± 0.102 when trained with a small dataset with dense expert annotations. However, extending the learning to larger databases would require prohibitive annotation efforts. Hence, in this work we implemented and compared new training methods that require only a small fraction of manually annotated pixels. While directly training on these scribble annotation deteriorates the segmentation quality by 26.8 percentage points, our new random walk-based semi-automatic target achieves the same Dice overlap as a dense supervision, and thus offers a more promising approach for sparse annotations.

A comparison of deep learning segmentation models for synchrotron radiation based tomograms of biodegradable bone implants

“…The study reported lower IoU values for the label degradation layer (79.31%-80.16%) than for the classes residual material and bone tissue, which achieved substantially higher IoU values (93.65%-93.10% and 96.72%-96.83%, respectively). Moreover, in a previous study using similar ex vivo data, Bockelmann et al [17] showed that the label "corroded screw" (degradation layer in this work) was the most challenging to be predicted, achieving a maximum Dice score of 54.1%. The Dice score is a metric similar to the IoU and also measures the performance of the semantic segmentation of ground truth data in comparison with predicted data.…”

“…However, in some cases, this step represents a major bottleneck, as mapping the structures into labels through the image greyscales is difficult with standard automated techniques [15]. In such cases, machine and deep learning algorithms, specifically convolutional neural networks (CNN) can be employed [10,[16][17][18][19][20][21][22].…”