BACKGROUND: This work aims to see whether Minkowski Functionals can be used to distinguish between cancer types before chemotherapy treatment has begun, and whether a response to treatment can be predicted by an initial scan alone. METHODS: Fat-nulled T1w 3T DCE-MRI scans were taken of 100 cases of biopsy confirmed breast cancer and a series of binary images created on lesion containing slices. Minkowski Functionals were calculated for each binary image and the change in these values as the binary threshold was raised was described using 6(th) order polynomials. These polynomials were used to compare between patient subgroups, for triple negative breast cancer (TNBC) status, chemotherapy response, biopsy grade, nodal status, and lymphovascular invasion status. RESULTS: When using Minkowski Functionals statistically significant (P < 0.05) differences were found between TNBC status, biopsy grade, and lymphovascular invasion status subgroups for all methodologies. The analysis performance did not appear to be affected by the number of threshold steps used. Most notably, very strong differences (P ≤ 0.01) were found between TNBC and other intrinsic subtype patients. When analyzed with a binary logistic regression model, an area under the curve value of 0.917 (0.846-0.987, 95% confidence interval) for TNBC classification was found. CONCLUSION: The method of texture analysis presented here provides a novel way to characterize tumors, and demonstrates clear differences between cancer groups which are detectable before treatment begins, and can help with treatment planning as a valuable prognosis tool.
BACKGROUND: This work aims to see whether Minkowski Functionals can be used to distinguish between cancer types before chemotherapy treatment has begun, and whether a response to treatment can be predicted by an initial scan alone. METHODS: Fat-nulled T1w 3T DCE-MRI scans were taken of 100 cases of biopsy confirmed breast cancer and a series of binary images created on lesion containing slices. Minkowski Functionals were calculated for each binary image and the change in these values as the binary threshold was raised was described using 6(th) order polynomials. These polynomials were used to compare between patient subgroups, for triple negative breast cancer (TNBC) status, chemotherapy response, biopsy grade, nodal status, and lymphovascular invasion status. RESULTS: When using Minkowski Functionals statistically significant (P < 0.05) differences were found between TNBC status, biopsy grade, and lymphovascular invasion status subgroups for all methodologies. The analysis performance did not appear to be affected by the number of threshold steps used. Most notably, very strong differences (P ≤ 0.01) were found between TNBC and other intrinsic subtype patients. When analyzed with a binary logistic regression model, an area under the curve value of 0.917 (0.846-0.987, 95% confidence interval) for TNBC classification was found. CONCLUSION: The method of texture analysis presented here provides a novel way to characterize tumors, and demonstrates clear differences between cancer groups which are detectable before treatment begins, and can help with treatment planning as a valuable prognosis tool.
Authors: Thibaud P Coroller; Vishesh Agrawal; Vivek Narayan; Ying Hou; Patrick Grossmann; Stephanie W Lee; Raymond H Mak; Hugo J W L Aerts Journal: Radiother Oncol Date: 2016-04-13 Impact factor: 6.280
Authors: Brittany Z Dashevsky; Jung Hun Oh; Aditya P Apte; Blanca Bernard-Davila; Elizabeth A Morris; Joseph O Deasy; Elizabeth J Sutton Journal: Sci Rep Date: 2018-01-10 Impact factor: 4.379
Authors: Stephan Ellmann; Evelyn Wenkel; Matthias Dietzel; Christian Bielowski; Sulaiman Vesal; Andreas Maier; Matthias Hammon; Rolf Janka; Peter A Fasching; Matthias W Beckmann; Rüdiger Schulz Wendtland; Michael Uder; Tobias Bäuerle Journal: PLoS One Date: 2020-01-30 Impact factor: 3.240