Computational Workflow for CE-SWI Image Processing

• 

  Fig 1: SWI image processing workflow developed in this study.
  Figure 1 presents the computational workflow for radiomic feature extraction developed in this study. CARPI was used to: 1) Batch-process the UPS contour files stored in the institutional network drive, 2) compute 107 radiomic features including 14 shape, 18 first-order statistics and 75 texture, and 3) automatically record them in a PostgreSQL-based relational database. 

CE-SWI Morphological Patterns in UPS

• 

  Fig 2: CE-SWI morphological patterns across treatment.
  Figure 2 shows a CE-SWI image of the right thigh demonstrating an undifferentiated pleomorphic sarcoma (UPS) at different stages during therapy: A-Baseline, B-Post-chemo initial, C-post chemo final, and D-Post-radiation with 99% necrosis on the histologic specimen.
• 

  Fig 3: Representative CE-SWI post-radiation morphology patterns.
  On Figure 3, (A-F) show 6 UPS cases demonstrating a T2* hypointense "complete ring“ pattern (green arrowheads). This is caused by hemosiderin and the deposition of fibrous tissue often observed in successfully treated sarcomas, with >90% TIN. (G-I) show 3 UPS cases with TIN in the range of 30-89% demonstrate an "incomplete ring" T2* hypointense pattern (yellow arrowheads pointing to areas of segmental ring absence). (J-L) show 3 UPS cases with <30% TIN showed a "globular" T2* hypointense pattern (red arrowheads).
• 

  Fig 4: Morphological patterns of CE-SWI at post-radiation by response.
  Figure 4 demonstrates that good responders displayed "complete ring" in 82%, partial responders displayed "incomplete ring" in 50%, and non-responders displayed "globular pattern" in 60%. Complete ring was significantly correlated with good responders (p=2.1x10-6), incomplete ring with partial responders (p=0.1562), and the globular pattern with non-responders (p=0.1455). 

High-Order Radiomics of CE-SWI: Complete Ring Pattern vs. Others

• 

  Fig 5: Complete hemosiderin ring as a UPS CE-SWI pattern of response.
  Figure 5 presents representative post-radiation CE-SWI images from 4 UPS patients demonstrating a “Complete Ring” pattern with a left-sided histogram (positive skewness).
• 

  Fig 6: Radiomics of CE-SWI ring morphology.
  Figure 6 shows that elongation, busyness, dependence variance, large area low gray level emphasis, gray level non-uniformity, zone variance, and large area emphasis provided statistically significant differences between "complete ring" pattern and others.

High-Order Radiomics of CE-SWI-CE: Good vs. Partial/Non-Responders

• 

  Fig 7: UPS CE-SWI radiomics at post-chemotherapy (PC).
  Figure 7 presents a scatter plot (A) with mean and 95% confidence interval for each of the 107 radiomic features extracted, and boxplots (B) of the top 10 features comparing good vs. partial/non-responders at post-chemotherapy. Dependence variance, 10th percentile, large dependence low gray level emphasis, low gray level emphasis, and dependence non-uniformity normalized provided statistically significant differences between good and partial/non-responders.
• 

  Fig 8: UPS CE-SWI radiomics at post-radiation (PRT).
  Figure 8 presents a scatter plot (A) with mean and 95% confidence interval for each of the 107 radiomic features extracted, and boxplots (B) of the top 10 features comparing good vs. partial/non-responders at post-radiation. Maximum probability and large area low gray level emphasis provided statistically significant differences between good and partial/non-responders.