Noise-robust foreground segmentation of multispectral imaging calibration volume in the presence of metallic implants for spectral range estimation in phantom and in-vivo data

Abstract

Purpose

Multispectral Imaging (MSI) methods can use a calibration scan to estimate an off-resonance field-map to determine the spectral range required to cover off-resonant signal in the presence of metallic implants of various shape and composition. Background signal noise can corrupt the field-map estimation in this calibration process. Previous work on foreground segmentation used a cumulative distribution function (CDF) to remove signal extrema, which can remove regions of true off-resonance signal from the calibration analysis.

The purpose of this work is to develop a foreground segmentation method robust to background noise in both phantom and in-vivo data to support calibrating the spectral range needed for MSI acquisitions.

Methods

The proposed method uses information from individual spectral bins, rather than a composite bin-combined image, for segmentation. Ten phantom (seven with metal) and ten in-vivo (six with metal) data were acquired using a prototype MSI spectral calibration sequence. Field-maps were estimated and spectral range estimates from the unmasked field-map and the proposed method were computed and compared using a paired sample Wilcoxon signed-rank test.

Results

The proposed method achieved a noise-robust foreground segmentation in both phantom and in-vivo data, in the presence or absence of metal devices. The Wilcoxon test showed a statistically significant difference between the spectral range estimates from the unmasked field-map and proposed method for both the phantom and in-vivo data (p-value: 0.002).

Conclusion

Noise-robust foreground segmentation achieved by the proposed method can improve the accuracy and robustness of spectral range estimates for time-efficient and reduced artifact multispectral imaging.