Building a pre-surgical multiparametric-MRI-based morphologic, qualitative, semiquantitative, first and high-order radiomic predictive treatment response model for undifferentiated pleomorphic sarcoma to replace RECIST

Figures

• 

  Fig. 1. Conventional sized-based response assessment metrics in responders and partial/non-responders comparing changes in diameter (RECIST), multiplication of the longest and perpendicular cross-sectional diameters (WHO), and volume at post-chemo (PC) and post-radiation (PRT) with respect to baseline (BL). The green and red vertical lines indicate partial response and progression thresholds, respectively, for RECIST, WHO, and volume. RECIST, WHO and Volume analysis all show universal pseudo progression at PC and stability at PRT, rendering these size-based metrics useless to separate R vs PR/NR

• 

  Fig. 2. Scatter plots showing means with 95% confidence intervals of ADC first-order radiomics at baseline (BL) vs. post-chemo (PC) vs. post-radiation (PRT) in responders and partial/non-responders. a.u.: Arbitrary Units. ADC mean and ADC skewness demonstrated the highest performance discriminating R vs PR/NR

• 

  Fig. 3. Frequency of CE-SWI morphological patterns, PWI/DCE morphological patterns, and PWI/DCE time-intensity curves (TICs) observed in responders, partial responders, and non-responders at post-radiation (PRT). Please refer to Tables 2–4 for examples of these patterns. CE-SWI Complete-Ring, PWI/DCE Capsular pattern, and PWI/DCE TIC type II are the most prevalent patterns observed in responders, while PWI/DCE Unipolar enhancement and PWI/DCE TIC type V are often seen in PR/NR

• 

  Fig. 4. Violin plots of seven PWI/DCE semi-quantitative parameters comparing responders (R) vs. partial/non-responders (PR/NR) at post-radiation (PRT). Significant differences (P < 0.05) were observed in WiR and WoR. a.u.: Arbitrary Units

• 

  Fig. 5. Violin plots of the top 10 CE-SWI radiomic features comparing responders (R) vs. partial/non-responders (PR/NR) at post-radiation (PRT). Statistically significant differences (P < 0.05) were observed in nine out of 10 radiomic features. s.v.: Standardized Value

• 

  Fig. 6. Receiver operating characteristic curves of classification models comparing responders vs. partial/non-responders at post-radiation based on (A) morphologic and qualitative features, (B) radiomic features, and (C) morphologic, qualitative, and radiomic features, (D) best performing DWI/ADC, PWI/DCE, and CE-SWI models, and (E) best-performing model overall. AUC: Area Under Curve

• 

  Fig. 7. Top panels: Representative UPS responder displaying a PWI/DCE type II curve (A), PWI/DCE Capsular pattern (B), and CE-SWI complete ring pattern (C) at PRT. Bottom panels: Representative UPS non-responder displaying a PWI/DCE type V curve (D), PWI/DCE bipolar pattern (E), and CE-SWI globular pattern (F) at PRT. PATE: Pathology-Assessed Treatment Effect