Automated quality assessment using appearance-based simulations and hippocampus segmentation on low-field paediatric brain MR images

TL;DR

The results show that although the images can provide understanding of large scale pathologies and gross scale anatomical development, there still remain barriers for their use for more granular analyses.

Abstract

Understanding the structural growth of paediatric brains is a key step in the identification of various neuro-developmental disorders. However, our knowledge is limited by many factors, including the lack of automated image analysis tools, especially in Low and Middle Income Countries from the lack of high field MR images available. Low-field systems are being increasingly explored in these countries, and, therefore, there is a need to develop automated image analysis tools for these images. In this work, as a preliminary step, we consider two tasks: 1) automated quality assurance and 2) hippocampal segmentation, where we compare multiple approaches. For the automated quality assurance task a DenseNet combined with appearance-based transformations for synthesising artefacts produced the best performance, with a weighted accuracy of 82.3%. For the segmentation task, registration of an average atlas performed the best, with a final Dice score of 0.61. Our results show that although the images can provide understanding of large scale pathologies and gross scale anatomical development, there still remain barriers for their use for more granular analyses.

Figures (5)

• Figure 1: Simulation of artefacts for various artefact domains for training the quality assessment models.
• Figure 2: Architecture of the multi-head decoder model for quality assessment.
• Figure 3: Selected ROI from T2
• Figure 4: Validation example segmentation results comparing the result from the four methods.
• Figure 5: Example hippocampal segmentations. Row 1: MNI T1 2mm template with the Harvard-Oxford probabilistic atlas, thresholded at 0.5. A): Example A showing a sample the 3D UNet repeatedly performed well (DSC $>$ 0.6) on with the tail of the atlas touching the ventricle. B) Example B showing a sample where the 3D UNet performed poorly (DSC $>$ 0.1), where the localisation of the hippocampus in the target appears different to the MNI template and sample A, with the label not reaching the ventricles. C) Example B with the average hippocampus mask registered to it, showing that the average hippocampus mask is located disjoint to the manual segmentation mask.