Liam Chalcroft

Development and evaluation of intraoperative ultrasound segmentation with negative image frames and multiple observer labels

L. Chalcroft, J. Qu, S.A. Martin, I.J.M.B. Gayo, G.V. Minore, I.R.D. Singh, S.U. Saeed, Q. Yang, Z.M.C. Baum, A. Altmann, Y. Hu

Advances in Simplifying Medical Ultrasound (ASMUS), MICCAI 2021 Workshop.

paperarXivcodeDOI

Abstract. When developing deep neural networks for segmenting intraoperative ultrasound images, several practical issues are encountered frequently, such as the presence of ultrasound frames that do not contain regions of interest and the high variance in ground-truth labels. In this study, we evaluate the utility of a pre-screening classification network prior to the segmentation network. Experimental results demonstrate that such a classifier, minimising frame classification errors, was able to directly impact the number of false positive and false negative frames. Importantly, the segmentation accuracy on the classifier-selected frames, that would be segmented, remains comparable to or better than those from standalone segmentation networks. Interestingly, the efficacy of the pre-screening classifier was affected by the sampling methods for training labels from multiple observers, a seemingly independent problem. We show experimentally that a previously proposed approach, combining random sampling and consensus labels, may need to be adapted to perform well in our application. Furthermore, this work aims to share practical experience in developing a machine learning application that assists highly variable interventional imaging for prostate cancer patients, to present robust and reproducible open-source implementations, and to report a set of comprehensive results and analysis comparing these practical, yet important, options in a real-world clinical application.

Published abstract

Year 2021Kind workshop paperVenue Simplifying Medical UltrasoundSeries Lecture Notes in Computer ScienceVolume 12967Pages 25-34DOI 10.1007/978-3-030-87583-1_3

Three trans-rectal ultrasound frames overlaid with prostate outlines from three observers in green, red and blue, showing close agreement, partial disagreement in extent, and disagreement on whether the prostate is present. Below, a flowchart in which each input frame passes through a classification network, and only frames scoring above a threshold reach the segmentation network; the rest are rejected.

Figure 1 from the author preprint (arXiv v1): observer labels and the pre-screening classifier workflow. The classifier decides whether a frame proceeds to segmentation.

Figure source

Open full-size figure

Materials

  • RectAngle implementation

    Training and evaluation code for ultrasound classification and segmentation, including pre-screening and the treatment of multiple observer labels.

  • Published chapter

    The ASMUS proceedings chapter, with the study methods and comparison of classifier-selected frames against standalone segmentation.

Cite

@inproceedings{chalcroft2021development,
  author    = {Chalcroft, Liam F. and Qu, Jiongqi and Martin, Sophie A. and Gayo, Iani JMB and Minore, Giulio V. and Singh, Imraj RD and Saeed, Shaheer U. and Yang, Qianye and Baum, Zachary M. C. and Altmann, Andre and Hu, Yipeng},
  title     = {{Development and evaluation of intraoperative ultrasound segmentation with negative image frames and multiple observer labels}},
  booktitle = {Simplifying Medical Ultrasound},
  year      = {2021},
  series    = {Lecture Notes in Computer Science},
  volume    = {12967},
  pages     = {25--34},
  publisher = {Springer},
  doi       = {10.1007/978-3-030-87583-1_3},
  url       = {https://link.springer.com/chapter/10.1007/978-3-030-87583-1_3}
}