Tumor shape can be used to non-invasively predict prognostic biomarkers of lower-grade gliomas

AJAS · 2022 Computer Science

Rohit Paradkar

Thumbnail supplied by the source for Tumor shape can be used to non-invasively predict prognostic biomarkers of lower-grade gliomas

Overview

Gliomas, which originate from glial cells, are considered the most aggressive type of brain tumors. Gliomas are categorized into four grades based on malignancy with lower-grade gliomas (LGG) being less severe than high grade gliomas (HGG). Current research efforts are focused mainly on HGGs, however LGGs are equally important research targets as they often develop into HGGs. This project aims to analyze MRIs containing LGGs via segmentation and utilize extracted tumor shape features to predict genomic subtypes. Segmentation, the process of outlining tumors in MRIs, is crucial to the diagnosis and treatment of a patient with an LGG. Generally, segmentation is performed manually by radiologists; however, this process is tedious, time-consuming, and often leads to inter-observer variability. Once an LGGs segmentation is complete, invasive biopsies and genetic tests such as RNAseq, miRNA sequencing, and RPPA, etc. are performed on the tumor cells. Recent studies suggest strong correlation between tumor shape and genomic subtypes stemming from the genetic tests. These relationships could be leveraged to devise a novel noninvasive method of identifying subtypes, which play a large role in diagnosis, treatment, and prognosis of LGGs. The specific aims of this project include analyzing LGGs through deep learning-based segmentation and shape feature extraction. Then, these shape features will be utilized to predict the genomic subtypes of the LGG. For automatic segmentation, two models were created using different convolutional neural networks (CNN). The highest performing model, which used a U-Net with a ResNeXt-50 backbone, yielded a 91.4% accuracy after testing. Shape feature extraction included computation of seven features which quantified tumor shape in 2D and 3D. Among the seven genetic tests of interest, six were shown to be correlated with shape features. These relationships were utilized in a deep-learning classifier which was capable of predicting genomic subtypes with accuracies between 60% and 80%. Altogether, this novel approach only requires patient MRIs for LGG segmentation and genomic subtype classification. Furthermore, this technology could increase time and cost efficiency, assist radiologists, and reduce the need for invasive diagnostic tests.

Video

If the video does not play, watch it on the original project page.

My Story

I was volunteering at the radiology department in my local hospital and saw MRIs of the brain. I remember asking the radiologist supervising me about them, and he told me about how MRI machines worked and the pulse sequences that were used for different contrasts. I was fascinated. Later, when trying to decide on a project related to lower-grade gliomas, I thought back to what I had learned at the radiology department and it struck me that I could utilize radiomic approaches to solve problems in glioma research. I worked out an experimental design based off the rationale that the shape of a tumor could be useful to determine diagnostic and prognostic information. I would later get the chance to become and presenter at the AJAS/AAS annual meeting and the rest is history!

From the student

References,

1) Wen, P. Y., & Reardon, D. A. (2016, January 18). Nature news. Retrieved February 21, 2021, from https://www.nature.com/articles/nrneurol.2015.242/figures/1

2) Stetka, B. (2019, March 27). New strategies take on the worst cancer--glioblastoma. Retrieved February 21, 2021, from https://www.scientificamerican.com/article/new-strategies-take-on-the-worst-cancer-glioblastoma/

3) Claus, E., Walsh, K., Wiencke, J., Molinaro, A., Wiemels, J., Schildkraut, J., . . . Wrensch, M. (2015, January). Survival and low-grade glioma: The emergence of genetic information. Retrieved February 21, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4361022/

4) Starmans, M. P., & Niessen, W. J. (2020). Manual segmentation. Retrieved February 21, 2021, from https://www.sciencedirect.com/topics/computer-science/manual-segmentation

5) Akkus, Z., Sedlar, J., Coufalova, L., Korfiatis, P., Kline, T., Warner, J., . . . Erickson, B. (2015, August 14). Semi-automated segmentation of pre-operative low grade gliomas in magnetic resonance imaging. Retrieved February 21, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4535671/

6) Buda, M., Saha, A., & Mazurowski, M. (2019, May 03). Association of genomic subtypes of lower-grade gliomas with shape features automatically extracted by a deep learning algorithm. Retrieved February 21, 2021, from https://www.sciencedirect.com/science/article/pii/S0010482519301520?casa_token=wMr02oVXdfwAAAAA%3AGcgbMv9MCXJl1R30MCr3zgQai-Nu10bULJLHkItwmNR-qQFBarGZHYsgtk6Bgq25zbkllZCcIPQ

7) Claus, E., Walsh, K., Wiencke, J., Molinaro, A., Wiemels, J., Schildkraut, J., . . . Wrensch, M. (2015, January). Survival and low-grade glioma: The emergence of genetic information. Retrieved April 03, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4361022/

8) The Cancer Genome Atlas Research Network*, Integrative Genomic Analysis of Diffuse Lower-Grade Gliomas (04:06), Author AffiliationsAddress reprint requests to Dr. Daniel J. Brat at the Department of Pathology and Laboratory Medicine, Integrative Genomic Analysis of Diffuse Lower-Grade Gliomas (04:06), Ellison, D., Others, J., . . . S. S. Abdool Karim and T. de Oliveira. (2015, June 25). Comprehensive, integrative genomic analysis of Diffuse LOWER-GRADE Gliomas: NEJM. Retrieved April 03, 2021, from https://www.nejm.org/doi/full/10.1056/nejmoa1402121

9) Chen, R., Smith-Cohn, M., Cohen, A., & Colman, H. (2017, April). Glioma Subclassifications and their clinical significance. Retrieved April 03, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5398991/

10) Brown, M., McDunn, J., Gunst, P., Smith, E., Milburn, M., Troyer, D., & Lawton, K. (2012, April 30). Cancer detection and biopsy classification using concurrent histopathological and metabolomic analysis of core biopsies. Retrieved April 03, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3446261/

11) Clark, K., Vendt, B., Smith, K., Freymann, J., Kirby, J., Koppel, P., . . . Prior, F. (2013, December 26). The cancer Imaging ARCHIVE (TCIA): Maintaining and operating a public information repository. Retrieved February 21, 2021, from https://pubmed.ncbi.nlm.nih.gov/23884657/

12) GDC. (n.d.). Retrieved February 21, 2021, from https://portal.gdc.cancer.gov/

13) Sankesara, H. (2019, January 23). U-net. Retrieved February 21, 2021, from https://towardsdatascience.com/u-net-b229b32b4a71

14) Xie, S., Girshick, R., Dollár, P., Tu, Z., & He, K. (2017, April 11). Aggregated Residual Transformations for Deep Neural Networks. Retrieved February 21, 2021, from https://arxiv.org/pdf/1611.05431.pdf

15) PyTorch. (n.d.). Retrieved February 21, 2021, from https://pytorch.org/hub/pytorch_vision_resnext/

16) Moshtagh, N. (2021). Minimum Volume Enclosing Ellipsoid. Retrieved February 21, 2021, from https://www.mathworks.com/matlabcentral/fileexchange/9542-minimum-volume-enclosing-ellipsoid

17) Imelfort, M. (2013, June 07). Minillinim/ellipsoid. Retrieved February 21, 2021, from https://github.com/minillinim/ellipsoid/blob/master/ellipsoid.py

18) Coelho, L., Ribeiro, R., Moreira, P., & Barbosa, M. (2019, April 10). Automatic Classification and Segmentation of Low-Grade Gliomas in Magnetic Resonance Imaging. Retrieved February 21, 2021, from https://doi.org/10.1007/978-3-030-17065-3_5

19) Ahmadvand, A., & Daliri, M. R. (2014, November 29). Brain mr image segmentation methods and applications. Retrieved February 21, 2021, from https://www.omicsonline.org/open-access/brain-mr-image-segmentation-methods-and-applications-2167-7964.1000e130.php?aid=33747

20) Johns Hopkins Medicine (Ed.). (n.d.). Gliomas. Retrieved February 21, 2021, from https://www.hopkinsmedicine.org/health/conditions-and-diseases/gliomas

21) Liu, Z., & Zhang, J. (2020, June 29). Radiogenomics correlation BETWEEN MR imaging features and mRNA-based subtypes IN lower-grade glioma. Retrieved April 03, 2021, from https://bmcneurol.biomedcentral.com/articles/10.1186/s12883-020-01838-6

22) Mazurowski, M. A., Clark, K., Czarnek, N. M., Shamsesfandabadi, P., Peters, K. B., & Saha, A. (2017, May 03). Radiogenomics of lower-grade glioma: Algorithmically-assessed TUMOR shape is associated with TUMOR genomic subtypes and patient outcomes in A multi-institutional study with the Cancer Genome Atlas data. Retrieved February 21, 2021, from https://link.springer.com/article/10.1007%2Fs11060-017-2420-1

23) Zhang, C., & Brat, D. J. (2016, January 12). Genomic profiling of lower-grade gliomas uncovers cohesive disease groups: Implications for diagnosis and treatment. Retrieved February 21, 2021, from https://cancercommun.biomedcentral.com/articles/10.1186/s40880-015-0071-1

24) Yale Medicine. (2020, November 17). Primary brain tumors. Yale Medicine. Retrieved January 23, 2022, from https://www.yalemedicine.org/conditions/primary-brain-tumors

Acknowledgments

Mandana Sassanfar, Director of Massachusetts Junior Academy of Science

Kirsten Vickey, Chemistry Teacher at Newton South High School