Abstract P6-01-06: Multi-Parametric MRI-Based Radiomics Models from Tumor and Peritumoral Regions as Potential Predictors of Treatment Response to Neoadjuvant Systemic Therapy in Triple Negative Breast Cancer Patients

Background and Purpose Early prediction of neoadjuvant systemic therapy (NAST) response in triple negative breast cancer (TNBC) patients could potentially aid in the selection of alternative therapies and avoid unnecessary toxicity in patients unlikely to achieve pathologic complete response (pCR) with NAST. In this study, we investigated the radiomic features of the peritumoral and the tumoral regions from dynamic contrast enhanced (DCE) MRI acquired at different time points of NAST for early treatment response prediction in TNBC. Methods and Materials This study included 182 biopsy-confirmed stage I-III TNBC patients enrolled in an IRB approved prospective clinical trial (NCT02276433). All patients underwent DCE-MRI on a GE 3T MRI scanner at baseline (BL), after two (C2) and four (C4) cycles of doxorubicin/cyclophosphamide based chemotherapy and before surgery. The peritumoral and the tumoral regions were segmented manually by two fellowship-trained radiologists using early phase (2.5 min) DCE-MRI subtraction images. Ten first order radiomic features, 300 grey-level-co-occurrence matrix (GLCM) features along with their absolute and relative differences (C4/BL, C2/BL, C4/C2) between the 3 imaging time points were extracted from the peritumoral and the tumoral regions. Patients were randomly divided into training and testing sets in a 2:1 ratio. For univariate analysis, area under the receiver operating characteristics curve (AUC ROC) was measured to determine the features most predictive of pCR/non-pCR. Wilcoxon Rank Sum test was used to test the statistical significance of predictive performance. In multivariate analysis, radiomic models were established using logistic regression with elastic net regularization followed by 5-fold cross validation for performance assessment. Results Eighty-eight (48%) patients had pCR (59 training, 29 testing) and 94 (52%) patients had non-pCR (63 training, 31 testing). Twenty-five radiomic features (4 from peritumoral C4, 5 from tumoral C4, 4 from peritumoral C4/BL, 6 from tumoral C4/BL, 2 from peritumoral C4/C2 and 4 from tumoral C4/C2) were statistically significant with AUC ≥ 0.75 in both the training and the testing sets at the univariate analysis. The significant features at C4 had AUCs of 0.75-0.79 for the training set and 0.76-0.81 for the testing set. Changes measured between C4 and BL or C2 showed AUC of 0.76-0.84 in the training and 0.75-0.81 in the testing datasets. Eleven multivariate regression models comprised of radiomic features at BL, C2, C4 and their changes (C4/BL, C4/C2 and C2/BL) showed an AUC of 0.80-0.84 for cross validation and an AUC of 0.80-0.82 for independent testing. Conclusions Radiomic models using longitudinal DCE MRI parameters of peritumoral and tumoral regions during NAST have the potential to predict pCR in TNBC patients undergoing NAST. Citation Format: Bikash Panthi, Rania M. Mohamed, Beatriz Adrada, Rosalind Candelaria, Mary S. Guirguis, Wei Yang, Medine Boge, Miral Patel, Nabil Elshafeey, Sanaz Pashapoor, Zijian Zhou, Jong Bum Son, Ken-Pin Hwang, H. T. Carisa Le-Petross, Jessica Leung, Marion E. Scoggins, Gary J. Whitman, Zhan Xu, Deanna L. Lane, Tanya Moseley, Frances Perez, Jason White, Elizabeth Ravenberg, Alyson Clayborn, Mark Pagel, Huiqin Chen, Jia Sun, Peng Wei, Alastair M. Thompson, Stacy Moulder, Anil Korkut, Lei Huo, Kelly K. Hunt, Jennifer K. Litton, Vicente Valero, Debu Tripathy, Clinton Yam, Jingfei Ma, Gaiane Rauch. Longitudinal DCE-MRI Radiomic Models for Early Prediction of Response to Neoadjuvant Systemic Therapy (NAST) in Triple Negative Breast Cancer (TNBC) Patients [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P6-01-34.

Background and Purpose Triple negative breast cancer (TNBC) is a biologically aggressive tumor and a refractory subtype of breast cancer due to the lack of therapeutic targets, such as estrogen receptors, progesterone receptors, and human epidermal growth factor receptor 2. In this study, we investigated the accuracy of radiomic models based on the dynamic contrast enhanced (DCE) MRI images obtained after the completion of NAST as discriminators of treatment response in TNBC patients. Materials and Methods This IRB-approved prospective study (ARTEMIS trial, NCT02276443) included 181 patients with biopsy proven stage I-III TNBC who Had MRIs after completion of NAST and before surgery. Patients were classified as pathologic complete response (pCR) and non-pCR at the surgery. Tumors were segmented on the 2.5 minutes DCE subtraction images. Regions with necrosis or clip artifacts were excluded from the contour. If tumors were not visible, the tumor bed was contoured. Whole-tumor histogram-based first order texture features (p=10) including mean, minimum, maximum, Standard deviation, kurtosis, skewness, 1st, 5th, 95th, and 99th percentiles, and radiomic (p=300) Grey Level Co-occurrence matrix (GLCM) features were extracted with an in-house Matlab toolbox. The samples were split into training and testing data sets by a 2:1 ratio. For univariate analysis area under the receiver operating characteristics curve (AUC ROC) was performed for pCR status prediction. For texture feature selection logistic regression with elastic net regularization was performed. Parameter optimization was performed by using 5-fold cross-validation based on mean cross-validated AUC in the training set. A P-value less than 0.05 was considered statistically significant. Results Of the total 181 patients, 88 (49%) had pCR and 93 (51%) had non-pCR. Univariate analysis identified 7 statistically significant first order imaging features (Minimum, Maximum, Mean, 1st Percentile, 5th Percentile, 95th Percentile, and 99th Percentile) with AUC >= 0.7 (p< 0.001), in both training and testing data sets. Percentile 5 showed highest AUC = 0.78 (p< 0.001). Two multivariate models were statistically significant at cross-validation with AUC>=0.7. The first model combined 2 first order data (Percentile 1 and Percentile 5) with AUC = 0.73 (p< 0.001). The second model combined 8 first order features (Percentile 1, 5, 95, 99, Mean, Minimum, Maximum, and Skewness) and 24 GLCM features with AUC = 0.7 (p=0.003). Conclusion DCE-MRI radiomic features from tumor and tumor bed regions in TNBC may be helpful imaging biomarkers for predicting treatment response after NAST. Citation Format: Rania M. Mohamed, Bikash Panthi, Beatriz Adrada, Rosalind Candelaria, Mary S. Guirguis, Wei Yang, Medine Boge, Miral Patel, Nabil Elshafeey, Sanaz Pashapoor, Zijian Zhou, Jong Bum Son, Ken-Pin Hwang, H. T. Carisa Le-Petross, Jessica Leung, Marion E. Scoggins, Gary J. Whitman, Zhan Xu, Deanna L. Lane, Tanya Moseley, Frances Perez, Jason White, Huiqin Chen, Jia Sun, Peng Wei, Jennifer K. Litton, Vicente Valero, Clinton Yam, Mark Pagel, Jingfei Ma, Gaiane Rauch. A Pre-operative Dynamic Contrast Enhanced MRI-Based Radiomics Models as Predictors of Treatment Response after Neoadjuvant Systemic Therapy in Triple Negative Breast Cancer Patients [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P6-01-35.

Background and Purpose: TNBC is comprised of biologically aggressive tumors with diverse clinical behavior and response to chemotherapy. Prediction of disease response to NACT is critical to the development of personalized medicine in TNBC. We evaluated first-order radiomic features from quantitative ADC maps of the tumor and peritumoral region as discriminators of response to NACT in TNBC patients. Materials and Methods: This IRB-approved prospective study (ARTEMIS trial, NCT02276443) included 34 patients with biopsy proven stage I-III TNBC who underwent evaluation of treatment response by multi-parametric MRI. Patients had a baseline MRI (BL) and a second MRI after 4 cycles (C4) of their treatment. After completion of NACT, all patients underwent surgery and were classified as pathologic complete response (pCR) or non-pCR. Both MRI exams included T2W series, a dynamic contrast enhanced series (DCE), a conventional diffusion weighted imaging (DWI) series, and a reduced field of view (rFOV) DWI series. Tumor volumes were contoured by an experienced breast radiologist on ADC maps with reference to b1000 DWI images. Regions with necrosis or clip artifacts were excluded from the contour. Peritumoral regions were defined as a 5 mm rim of tissue surrounding the tumor based on DCE series, T2-weighted images with fat suppression and ADC maps. Thirteen first-order radiomic features, including mean, minimum, maximum, percentiles, kurtosis and skewness at a single measurement and the difference between BL and C4 were compared between pCR and non-pCR using Receiver Operating Characteristic (ROC) curve and Wilcoxon rank sum test. Results: The kurtosis of tumor at C4 by conventional DWI was significantly higher in non-pCR than in pCR patients (AUC=0.785, p=0.0097). The change in kurtosis from BL to C4 by conventional DWI was also significantly higher in non-pCR than in pCR patients (AUC=0.73, p=0.043). The skewness of tumor at C4 by rFOV DWI scan was significantly lower in pCR than non-pCR patients (AUC=0.73, p=0.023). The 10th percentile of the peritumoral region’s ADC was significantly different between pCR and non-pCR (mean=1.19, SD is ± 0.27 10-3 mm2/s vs mean=1.34, SD ± 0.27 10-3 mm2/s respectively, AUC=0.70, p=0.048). The kurtosis and 25th percentile of the ADC of peritumoral region were borderline significantly different between pCR and non-pCR (AUC=0.69, p=0.067; AUC=0.69, p= 0.073 respectively). Conclusion: ADC first-order radiomic features from tumor and peritumoral region in TNBC may be useful for predicting treatment response to NACT. Larger study is necessary and is currently in progress to validate these findings. Citation Format: Beatriz E. Adrada, Abeer H. Abdelhafez, Benjamin C. Musall, Kenneth R. Hess, Jong Bum Son, Mark D. Pagel, Ken-Pin Hwang, Rosalind P. Candelaria, Lumarie Santiago, Gary J. Whitman, Huong Le-Petross, Tanya W. Moseley, Elsa Arribas, Deanna L. Lane, Marion E. Scoggins, David A. Spak, Jessica W.T. Leung, Senthil Damodaran, Bora Lim, Vicente Valeo, Jason B White, Alastair M. Thompson, Jennifer K. Litton, Stacy L. Moulder, Jingfei Ma, Wei T. Yang, Gaiane M Rauch. Quantitative apparent diffusion coefficient (ADC) radiomics of tumor and peritumoral regions as potential predictors of treatment response to neoadjuvant chemotherapy (NACT) in triple negative breast cancer (TNBC) patients [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr P6-02-03.

Background and Purpose:Early and accurate assessment ofbreast cancer response to NAST is important for patient management. In this study, we investigated the value of radiomic phenotypes derived from semi-quantitative and quantitative DCE-MRI parametric maps for early prediction of NASTresponse in TNBC patients. MATERIALS AND METHODS:This IRB approved study included 74 patients with stage I-III TNBC who were enrolled in the prospective ARTEMIS trial (NCT02276443). Pathologic complete response (pCR) and non-pCR were assessed by surgical histopathology after NAST (pCR=34; non-pCR=40).MRI scans were obtained at 3 time points during the NAST treatment with every 2-week anthracycline-based chemotherapy (AC): at baseline (BSL=74), post-2 cycles of AC (C2= 27) and post-4 cycles of AC (C4= 27). Patients went on to receive taxane-based chemotherapy prior to surgery. Tumor regions of interest (ROIs) were segmented by a breast radiologist at the early-phase subtractions of DCE-MRI scans using in-house developed software, followed by co-registration of the ROIs with quantitative (Ktrans, Veand Kep), and semi-quantitative DCE parametric maps (Maximum Slope Increase (MSI), Positive Enhancement Integral (PEI) and Peak Signal Enhancement Ratio (SER)).A total of 93 first order radiomic features were extracted from the tumor ROIs of each time point semi-quantitative DCE parametric map, while a total of 390 extracted radiomic features (first order-histogram features and second order grey-level-co-occurrence matrix) were extracted from each quantitative DCE parametric map using an in-house developed Matlab software.Radiomic features at each time point and changes between the 3 time points were compared between pCR and non-pCR using Wilcoxon Rank Sum test and Fisher’s exact test. Area under the receiver operating characteristics curve (AUC) was used to determine which features predicted pCR.Logistic regression was performed for feature selection, and used to build the radiomic phenotype model. The model performance was assessed by leave-one-out cross validation and 3-fold cross validation. RESULTS:Thirty-three radiomic features from PEI map were significantly different between pCR and non-pCR. The PEI most significant features were changesbetween BSL and C4 in skewness, mean and median (AUC=0.87, 0.85 and 0.87, p=<0.001, 0.001 and 0.002 respectively). Additionally, 31 MSI features were significantly different between pCR and non-pCR. The top 2 features were the interscan-change in skewness between BSL and C2 (AUC=0.80, P=0.007) and C4 standard deviation (AUC=0.80, P=0.006). Four BSL Veradiomic features were statistically significant between pCR and non-pCR with the best being range of difference variance (AUC=0.64, P=0.03). One BSL Kepfeature (Angular-Variance of Information measure of correlation-2) was able to differentiate pCR from non-pCR (AUC=0.64, P=0.04). Five C4-Ktrans features were able to differentiate pCR and non-pCR, with the most significant being mean value (AUC=0.86, P=0.001). BSL-Kepradiomic model built from 24 features (AUC=0.80, p=0.003) and combined (Ktrans, Veand Kep)C2-radiomic model consisting of 20 features (AUC=0.97, p=0.01) showed the best performance for prediction of pCR. CONCLUSIONS:Radiomic phenotypes form DCE-MRI parametric maps were useful for differentiation between pCR and non-pCR and showed promise as noninvasive imaging biomarkers for early prediction of NAST response in TNBC. Potentially, DCE-MRI radiomic features may be used for development of diagnostic predictive model for early noninvasive assessment of NAST treatment response in TNBC patients. Citation Format: Nabil Elshafeey, Beatriz E Adrada, Rosalind P Candelaria, Abeer H Abdelhafez, Benjamin C Musall, Jia Sun, Medine Boge, Rania M.M Mohamed, Hagar S Mahmoud, Jong Bum Son, Aikaterini Kotrosou, Shu Zhang, Jessica Leung, Deanna Lane, Marion Scoggins, David Spak, Elsa Arribas, Lumarie Santiago, Gary J. Whitman, Huong T Le-Petross, Tanya W Moseley, Jason B White, Elizabeth Ravenberg, Ken-Pin Hwang, Peng Wei, Jennifer K Litton, Lei Huo, Debu Tripathy, Vicente Valero, Alastair M Thompson, Stacy Moulder, Wei T Yang, Mark D Pagel, Jingfei Ma, Gaiane M Rauch. Radiomic phenotypes from dynamic contrast-enhanced MRI (DCE-MRI) parametric maps for early prediction of response to neoadjuvant systemic therapy (NAST) in triple negative breast cancer (TNBC) patients [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PD6-06.

Background and Purpose:There is currently a lack of recognized imaging criteria for prediction of treatment response to NAST in breast cancer patients with recent reports showing that breast MRI is the most accurate modality for evaluation of NAST response. DCE-MRI evaluates tumor perfusion that influences tumor enhancement at the post-contrast subtraction images and allows for more accurate measurement of changes in tumor volume during NAST. In this study, we evaluated the ability of tumor volumetric changes after 2 and 4 cycles of NAST by longitudinal ultrafast DCE-MRI to predict pathologic complete response (pCR) in TNBC undergoing NAST. Materials and Methods: Stage I-III TNBC patients enrolled in an IRB approved prospective clinical trial (ARTEMIS, NCT02276433) who had ultrafast DCE-MRI at baseline (BL, N=103), post 2 cycles (C2, N=59), and post 4 cycles (C4, N=103) of anthracycline-based NAST,and had surgery, were included in this analysis. Tumor volume was calculated using 3D measurements of the index lesion at BL, C2, and C4. Percent change of tumor volume (%TV) between BL, C2, and C4 was calculated at early (9-12 sec) and delayed (360-480 sec) phases of DCE-MRI. The largest lesion was used for analysis in patients with multicentric or multifocal disease. Demographic, clinical, and pathologic data and treatment response at surgery (pCR versus non-pCR) were documented. Receiver operating characteristics curve (ROC) analysis was performed for prediction of pCR status. Positive predictive value (PPV), negative predictive value (NPV) and Youden Index were used to select %TV cut-off thresholds for pCR prediction.Results: 103 patients (median age, 53 years; range, 24-79 years) were included, 48 (47%) had pCR, and 55 (53%) had non-pCR at surgical pathology. The %TV reduction at C2 DCE-MRI was predictive of pCR on both early phase DCE MRI (AUC, 0.873; CI:0.779-0.968, p < .0001) and delayed phase DCE MRI (AUC, 0.844; CI:0.742-0.947, p < .0001). Optimal thresholds were as follows: 70% TV reduction on early phase DCE MRI with Youden’s index of 1.58 was able to predict pCR correctly for 79% of patients with PPV of 81%; 75% TV reduction on delayed phase with Youden’s Index of 1.44 was able to predict pCR correctly for 71% of patients with PPV of 85%.%TV reduction was also predictive of pCR at the C4 time point on both early phase DCE MRI (AUC, 0.761; CI:0.665-0.856, p < .0001) and delayed phase DCE MRI (AUC, 0.737; CI:0.641-0.833, p < .0001). Optimal thresholds were as follows: 90% TV reduction on early phase DCE MRI with Youden’s index of 1.43 was able to correctly predict pCR in 72% of patients with PPV of 70%; and 90% TV reduction on delayed phase with Youden’s Index of 1.34 was able to predict pCR correctly in 68% of patients with PPV of 71%.Conclusion: Our data shows that percent tumor volume reduction by DCE-MRI after 2 and 4 cycles of NAST was able to predict pCR in TNBC with high accuracy and can be used as an early imaging biomarker of NAST response prediction. Volumetric changes by longitudinal DCE-MRI can be used to differentiate chemoresistant and chemosensitive TNBC patients as early as after 2 cycles of NAST, and can help to triage patients for treatment de-escalation or targeted therapy. Citation Format: Gaiane Margishvili Rauch, Adrada E Beatriz, Rosalind P Candelaria, Nabil Elshafeey, Abeer H Abdelhafez, Benjamin C Musall, Jia Sun, Medina Boge, Rania M.M Mohamed, Jong Bum Son, Shu Zhang, Jessica Leung, Deanna Lane, Marion Scoggins, David Spak, Elsa Arribas, Lumarie Santiago, Gary J Whitman, Huong T. Le-Petross, Tanya W Moseley, Jason B. White, Elizabeth Ravenberg, Ken-Pin Hwang, Peng Wei, Lei Huo, Jennifer K Litton, Vicente Valero, Debu Tripathy, Alastair M Thompson, Mark D Pagel, Jingfei Ma, Wei T Yang, Stacy Moulder. Volumetric changes on longitudinal dynamic contrast enhanced MR imaging (DCE-MRI) as an early treatment response predictor to neoadjuvant systemic therapy (NAST) in triple negative breast cancer (TNBC) patients [abstract]. In: Proceedings of the 2020 San Antonio Breast Cancer Virtual Symposium; 2020 Dec 8-11; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2021;81(4 Suppl):Abstract nr PD6-07.

Early prediction of neoadjuvant systemic therapy (NAST) response for triple-negative breast cancer (TNBC) patients could help oncologists select individualized treatment and avoid toxic effects associated with ineffective therapy in patients unlikely to achieve pathologic complete response (pCR). The objective of this study is to evaluate the performance of radiomic features of the peritumoral and tumoral regions from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) acquired at different time points of NAST for early treatment response prediction in TNBC. This study included 163 Stage I-III patients with TNBC undergoing NAST as part of a prospective clinical trial (NCT02276443). Peritumoral and tumoral regions of interest were segmented on DCE images at baseline (BL) and after two (C2) and four (C4) cycles of NAST. Ten first-order (FO) radiomic features and 300 gray-level-co-occurrence matrix (GLCM) features were calculated. Area under the receiver operating characteristic curve (AUC) and Wilcoxon rank sum test were used to determine the most predictive features. Multivariate logistic regression models were used for performance assessment. Pearson correlation was used to assess intrareader and interreader variability. Seventy-eight patients (48%) had pCR (52 training, 26 testing), and 85 (52%) had non-pCR (57 training, 28 testing). Forty-six radiomic features had AUC at least 0.70, and 13 multivariate models had AUC at least 0.75 for training and testing sets. The Pearson correlation showed significant correlation between readers. In conclusion, Radiomic features from DCE-MRI are useful for differentiating pCR and non-pCR. Similarly, predictive radiomic models based on these features can improve early noninvasive treatment response prediction in TNBC patients undergoing NAST.

To predict the response of breast cancer patients to neoadjuvant chemotherapy (NAC) using features derived from dynamic contrast-enhanced (DCE) MRI. 60 patients with triple-negative early-stage breast cancer receiving NAC were evaluated. Features assessed included clinical data, patterns of tumor response to treatment determined by DCE-MRI, MRI breast imaging-reporting and data system descriptors, and quantitative lesion kinetic texture derived from the gray-level co-occurrence matrix (GLCM). All features except for patterns of response were derived before chemotherapy; GLCM features were determined before and after chemotherapy. Treatment response was defined by the presence of residual invasive tumor and/or positive lymph nodes after chemotherapy. Statistical modeling was performed using Lasso logistic regression. Pre-chemotherapy imaging features predicted all measures of response except for residual tumor. Feature sets varied in effectiveness at predicting different definitions of treatment response, but in general, pre-chemotherapy imaging features were able to predict pathological complete response with area under the curve (AUC)=0.68, residual lymph node metastases with AUC=0.84 and residual tumor with lymph node metastases with AUC=0.83. Imaging features assessed after chemotherapy yielded significantly improved model performance over those assessed before chemotherapy for predicting residual tumor, but no other outcomes. DCE-MRI features can be used to predict whether triple-negative breast cancer patients will respond to NAC. Models such as the ones presented could help to identify patients not likely to respond to treatment and to direct them towards alternative therapies.

Pathologic complete response (pCR) to neoadjuvant systemic therapy (NAST) in triple-negative breast cancer (TNBC) is a strong predictor of patient survival. Edema in the peritumoral region (PTR) has been reported to be a negative prognostic factor in TNBC. To determine whether quantitative apparent diffusion coefficient (ADC) features from PTRs on reduced field-of-view (rFOV) diffusion-weighted imaging (DWI) predict the response to NAST in TNBC. Prospective. A total of 108 patients with biopsy-proven TNBC who underwent NAST and definitive surgery during 2015-2020. A 3.0 T/rFOV single-shot diffusion-weighted echo-planar imaging sequence (DWI). Three scans were acquired longitudinally (pretreatment, after two cycles of NAST, and after four cycles of NAST). For each scan, 11 ADC histogram features (minimum, maximum, mean, median, standard deviation, kurtosis, skewness and 10th, 25th, 75th, and 90th percentiles) were extracted from tumors and from PTRs of 5 mm, 10 mm, 15 mm, and 20 mm in thickness with inclusion and exclusion of fat-dominant pixels. ADC features were tested for prediction of pCR, both individually using Mann-Whitney U test and area under the receiver operating characteristic curve (AUC), and in combination in multivariable models with k-fold cross-validation. A P value < 0.05 was considered statistically significant. Fifty-one patients (47%) had pCR. Maximum ADC from PTR, measured after two and four cycles of NAST, was significantly higher in pCR patients (2.8 ± 0.69 vs 3.5 ± 0.94 mm2 /sec). The top-performing feature for prediction of pCR was the maximum ADC from the 5-mm fat-inclusive PTR after cycle 4 of NAST (AUC: 0.74; 95% confidence interval: 0.64, 0.84). Multivariable models of ADC features performed similarly for fat-inclusive and fat-exclusive PTRs, with AUCs ranging from 0.68 to 0.72 for the cycle 2 and cycle 4 scans. Quantitative ADC features from PTRs may serve as early predictors of the response to NAST in TNBC. 1 TECHNICAL EFFICACY: Stage 4.

Background and Purpose: There is currently lack of recognized imaging criteria for prediction of treatment response to NAST in breast cancer patients. And early identification of treatment response to neoadjuvant systemic therapy (NAST) in Triple Negative Breast Cancer (TNBC) patients is important for appropriate treatment selection and response monitoring. A novel MRI sequence, Magnetic Resonance Image Compilation (MagIC) is capable of simultaneous quantitation of several tissue water properties including longitudinal (T1), transverse (T2) relaxation times, and proton density (PD). In this study we evaluated the ability of a radiomic model extracted from a novel MagIC sequence acquired early during NAST to predict pathologic complete response to NAST in TNBC. Materials and Methods: This IRB approved prospective ARTEMIS trial (NCT02276443) included 184 women (122 training dataset, 62 testing dataset) diagnosed with stage I-III TNBC. All patients were scanned with MagIC on a 3T MRI scanner at baseline (184 patients), and after 4 cycles (156 Patients) of NAST. T1, T2 and PD maps were generated from the source images using SyMRI (SyntheticMR, Linkoping, Sweden). Histopathology at surgery was used to determine pathologic complete response (pCR) which was defined as absence of the invasive cancer in the breast and axillary lymph nodes. 3D contouring of the tumors was performed using an in-house toolbox. 310 (10 first-order, 300 GLCM) textural features were extracted from each map, with total of 930 features/patient. Radiomic features were compared between pCR and non-pCR using Wilcoxon Rank Sum test and Fisher’s exact test. To build a multivariate, predictive model, logistic regression with elastic net regularization was performed for texture feature selection. The tuning parameter was optimized using 5-fold cross-validation based on the average area under curve (AUC) of each fold of a cross-validation using training data. Then the testing data were used to compare model’s performance by AUC. Results: Univariate analysis found 23 PD, 17 T1 and 10 T2 radiomic features at C4 time point to be able to predict pCR status with AUC &amp;gt;70% in both training and testing cohort. The top performing radiomic features were Entropy, Variance, Homogeneity and Energy (Tables1-2). Multivariate radiomics models from C4-PD, and C4-T1 maps showed best performance during both cross validation and independent testing. The radiomic signature of C4-T1 map that included 27features had best performance, with an AUC of 0.77, 0.70 (95% CI: 0.571-0.868) in training and testing cohort respectively. C4-PD map radiomic signature that included 6features was able to predict the pCR status with AUC of 0.73, 0.72 (95% CI: 0.571-0.868) in training and testing cohort respectively. Conclusion: Our data found that MagIC-based radiomics signature could potentially predict pathologic complete response in TNBC early during NAST. This data shows the potential application of MagIC radiomic model for improvement of response assessment in TNBC. Table 1.Best performing radiomic features from PD map after 4 cycles of NAST in TNBC patients.FeatureTraining CohortTraining CohortTraining CohortTesting CohortTesting CohortTesting CohortNAUC95% CINAUC95% CIP-valuePD-mapAngular Variance of Sum entropy1060.73820.6437-0.8328500.73240.5895-0.8752&amp;lt;0.001Range of Sum entropy1060.73930.6446-0.834500.72120.5753-0.867&amp;lt;0.001Angular Variance of Sum entropy1060.75960.6662-0.853500.70190.5538-0.8501&amp;lt;0.001Average of Sum entropy1060.73470.6367-0.8327500.70990.5613-0.8585&amp;lt;0.001Angular Variance of Sum variance1060.70160.602-0.8011500.70190.5543-0.8495&amp;lt;0.001Range of Sum variance1060.70050.6001-0.8009500.700.5499-0.8476&amp;lt;0.001 Table 2.Best performing radiomic features from T1-T2 maps after 4 cycles of NAST in TNBC patients.FeatureTraining CohortTraining CohortTraining CohortTesting CohortTesting CohortTesting CohortNAUC95% CINAUC95% CIP-valueT1-mapAngular Variance of Sum entropy1060.76530.6762-0.8544500.70510.5524-0.8579&amp;lt;0.001Range of Sum entropy1060.76530.6759-0.8547500.70350.5503-0.8567&amp;lt;0.001Average of Entropy1060.75250.6568-0.8482500.71630.572-0.8607&amp;lt;0.001Average of Sum entropy1060.750.6552-0.8448500.70190.555-0.8488&amp;lt;0.001Angular Variance of Energy1060.7450.6493-0.8407500.73080.59-0.8715&amp;lt;0.001Range of Energy1060.74290.6466-0.8392500.72920.5885-0.8699&amp;lt;0.001Average of Energy1060.74110.6438-0.8384500.7260.5852-0.8667&amp;lt;0.001Average of Entropy1060.73360.635-0.8322500.74040.602-0.8787&amp;lt;0.001Average of Maximum probability1060.70760.6054-0.8098500.71630.5704-0.8623&amp;lt;0.001Range of Maximum probability1060.70550.6018-0.8092500.75640.6195-0.8933&amp;lt;0.001T2-mapAngular Variance of Energy1060.74820.6531-0.8433500.70990.5644-0.8555&amp;lt;0.001Range of Energy1060.7450.6495-0.8405500.70350.5569-0.8501&amp;lt;0.001Average of Entropy1060.74070.6416-0.8399500.72920.585-0.8733&amp;lt;0.001Average of Sum entropy1060.73860.6405-0.8367500.72440.5797-0.869&amp;lt;0.001Average of Energy1060.73180.6309-0.8327500.72120.5743-0.86&amp;lt;0.001Angular Variance of Sum entropy1060.7290.631-0.827500.72760.5857-0.8695&amp;lt;0.001Range of Sum entropy1060.72760.6295-0.8257500.72280.5796-0.8659&amp;lt;0.001Average of Information measure of correlation 11060.71580.6147-0.8169500.70990.5638-0.8561&amp;lt;0.001Average of Entropy1060.700.5903-0.8028500.74360.6014-0.8858&amp;lt;0.001 Citation Format: Nabil Elshafeey, Ken-Pin Hwang, Beatriz Elena Adrada, Rosalind Pitpitan Candelaria, Medine Boge, Rania M Mahmoud, Huiqin Chen, Jia Sun, Wei Yang, Aikaterini Kotrotsou, Benjamin C Musall, Jong Bum Son, Gary J Whitman, Jessica Leung, Huong Le-Petross, Lumarie Santiago, Deanna Lynn Lane, Marion Elizabeth Scoggins, David Allen Spak, Mary Saber Guirguis, Miral Mahesh Patel, Frances Perez, Abeer H Abdelhafez, Jason B White, Lei Huo, Elizabeth Ravenberg, Wei Peng, Alastair Thompson, Senthil Damodaran, Debu Tripathy, Stacey L Moulder, Clinton Yam, Mark David Pagel, Jingfei Ma, Gaiane Margishvili Rauch. Radiomics model based on magnetic resonance image compilation (MagIC) as early predictor of pathologic complete response to neoadjuvant systemic therapy in triple-negative breast cancer [abstract]. In: Proceedings of the 2021 San Antonio Breast Cancer Symposium; 2021 Dec 7-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2022;82(4 Suppl):Abstract nr PD11-06.

Diffusion tensor imaging (DTI) is an advanced diffusion MRI model designed to extend the spatial microstructural information achieved by diffusion-weighted imaging (DWI). By acquiring at least six directional diffusion gradients and modeling the respective diffusion coefficients, DTI can extract multiple parameters that characterize the diffusion process and its directional dependence (anisotropy) at pixel resolution. The motivation to apply DTI for breast imaging arises from the ductal tree shape of the mammary gland and the assumed intraductal anisotropic diffusivity.1 Indeed, DTI provided quantified characterization for the microstructural transformations that the breast experiences during lactation and weaning.2, 3 Clinically, breast DTI parametric maps were found to be useful in breast cancer detection,4 as well as in advanced imaging of pregnancy-associated breast cancer,5 by providing an unenhanced alternative to dynamic contrast-enhanced (DCE) MRI during pregnancy and increased tumor conspicuity vs. DCE during lactation.6 Neoadjuvant systemic therapy (NST) is an evolving breast cancer management approach, which refers to the preoperative administration of chemo and endocrine therapy, mainly in patients with locally advanced or metastatic disease, with the aim of down-staging and breast surgery optimization.7 Another advantage of NST lies in its ability to identify patients with pathologic complete response (pCR) preoperatively. pCR is an important predictive marker for long-term favorable outcomes, especially among patients with triple-negative breast cancer (TNBC) and can guide further escalation or de-escalation strategies in the adjuvant setting. While the gold standard for PCR is determined pathologically rather than radiologically, several MRI sequences, including DTI, were shown to be useful in monitoring response to NST.8, 9 In the present prospective clinical trial,10 86 TNBC patients were enrolled to undergo consecutive DTI examinations at three time points in the setting of NST. Longitudinal changes in DTI-derived parameters were measured within the tumor, peri-tumoral region (PTR), and the ipsilateral breast fibroglandular tissue (FGT) and tested for discriminating between patients with and without pCR. Results showed a significant difference between complete responders (N = 40) and non-pCR patients for several DTI parameters, including the median surface/average anisotropy of the PTR, measured after two and four cycles of NAST, which increased in pCR patients and decreased in non-pCR patients (P < 0.001, area under the curve [AUC] 0.78) and FGT at baseline and cycle four scans (AUC 0.64). While the separation between the subgroups was shown to be modest, and overlapping values were encountered for the various parameters, this study by Musall et al10 highlights the potential of multi-parametric MRI and DTI to add valuable diagnostic information on breast cancer response to treatment. In addition, the abovementioned study also stresses the importance of peri-tumoral and ipsilateral breast characterization rather than focusing solely on the tumor itself. It remains to be seen, though, whether there are specific parameters (even at the expense of limited sensitivity) that could identify complete responders with high confidence. This would signify a step forward in the efficacy of radiological pCR evaluation in comparison with contemporary references and would enable the utilization of MRI as a viable alternative to histopathology determined pCR. In summary, peri-tumoral longitudinal changes in DTI-derived parameters, investigated by Musall et al,10 were able to predict NST response among TNBC patients. Further validation and research are required to recognize highly specific parameters that can propel its clinical integration and offer a reliable alternative for surgical pCR.

The purpose of the present study was to examine the potential of a machine learning model with integrated clinical and CT-based radiomics features in predicting pathologic complete response (pCR) to neoadjuvant systemic therapy (NST) in breast cancer. Contrast-enhanced CT was performed in 329 patients with breast tumors (n = 331) before NST. Pyradiomics was used for feature extraction, and 107 features of seven classes were extracted. Feature selection was performed on the basis of the intraclass correlation coefficient (ICC), and six ICC thresholds (0.7-0.95) were examined to identify the feature set resulting in optimal model performance. Clinical factors, such as age, clinical stage, cancer cell type, and cell surface receptors, were used for prediction. We tried six machine learning algorithms, and clinical, radiomics, and clinical-radiomics models were trained for each algorithm. Radiomics and clinical-radiomics models with gray level co-occurrence matrix (GLCM) features only were also built for comparison. The linear support vector machine (SVM) regression model trained with radiomics features of ICC ≥0.85 in combination with clinical factors performed the best (AUC = 0.87). The performance of the clinical and radiomics linear SVM models showed statistically significant difference after correction for multiple comparisons (AUC = 0.69 vs. 0.78; p &lt; 0.001). The AUC of the radiomics model trained with GLCM features was significantly lower than that of the radiomics model trained with all seven classes of radiomics features (AUC = 0.85 vs. 0.87; p = 0.011). Integration of clinical and CT-based radiomics features was helpful in the pretreatment prediction of pCR to NST in breast cancer.

Simple SummaryNeoadjuvant systemic therapy (NAST) is given before surgery to reduce tumor burden in patients with triple-negative breast cancer (TNBC), which is an aggressive breast cancer subtype that accounts for approximately 30% of breast cancer-related mortalities. Unfortunately, approximately 50% of TNBC patients do not respond to NAST and develop distant spread within 5 years. Reliable clinical methods are needed to determine non-responders to NAST in order to avoid the severe toxicity of ineffective regimens and offer novel targeted treatments. The purpose of this study was to investigate functional tumor volume measured from dynamic contrast-enhanced MRI for early assessment of NAST response in TNBC. Our study demonstrated the potential of functional tumor volume, evaluated as early as after 2 and 4 cycles of NAST, to serve as a non-invasive biomarker for the prediction of treatment response in TNBC patients.Early assessment of neoadjuvant systemic therapy (NAST) response for triple-negative breast cancer (TNBC) is critical for patient care in order to avoid the unnecessary toxicity of an ineffective treatment. We assessed functional tumor volumes (FTVs) from dynamic contrast-enhanced (DCE) MRI after 2 cycles (C2) and 4 cycles (C4) of NAST as predictors of response in TNBC. A group of 100 patients with stage I-III TNBC who underwent DCE MRI at baseline, C2, and C4 were included in this study. Tumors were segmented on DCE images of 1 min and 2.5 min post-injection. FTVs were measured using the optimized percentage enhancement (PE) and signal enhancement ratio (SER) thresholds. The Mann–Whitney test was used to compare the performance of the FTVs at C2 and C4. Of the 100 patients, 49 (49%) had a pathologic complete response (pCR) and 51 (51%) had a non-pCR. The maximum area under the receiving operating characteristic curve (AUC) for predicting the treatment response was 0.84 (p < 0.001) for FTV at C4 followed by FTV at C2 (AUC = 0.82, p < 0.001). The FTV measured at baseline was not able to discriminate pCR from non-pCR. FTVs measured on DCE MRI at C2, as well as at C4, of NAST can potentially predict pCR and non-pCR in TNBC patients.

Background: TNBC patients exhibit an aggressive disease course with at least 25% developing recurrence or metastasis within 3-5 years, despite standard-of-care therapies. Clinicopathologic factors (CP; age, growth pattern, tumor size, margin status and grade) have limited value in identifying high risk TNBC patients. Early and accurate prediction of recurrence in TNBC patients from clinical mammograms would facilitate therapy optimization. However, new strategies to identify high-risk patients are needed. In this study, we used radiomics features at the tumor boundary to predict recurrence in TNBC patients and LLM/GPT to explain the prediction in natural language. Materials and methods: Mammograms from node negative TNBC patients (n=77; aged 30-90 years, lesion size of 2-45 mm, grades 2 and 3) who underwent adjuvant chemotherapy and had 5-yr follow-up were analyzed. Lesions were manually segmented and tumor boundary (1 mm containing both intra-tumoral and peritumoral regions) was automatically determined. Over 2,000 radiomics features from the tumor boundary and the central regions of the tumor were extracted to quantify heterogeneity, texture, shape, and size of the tumor. Recursive feature elimination using non-linear random forest classifier was used to reduce feature dimensionality and prevent over-fitting. A random forest classifier was employed to build an AI/machine learning model for risk of recurrence using the reduced feature dimension. Top ranking features along with clinicopathological variables were further used in a Bayesian network (BN) to create an interpretable machine learning model. To generate a patient-specific report, a sentence transformer model was further used for BN graph embedding, which aided in querying the graph in natural language inheriting the node/feature specific dependencies of the BN in LLM/GPT (BN LLM). Results: Recurrence prediction was significantly improved by analysis of radiomics features at the tumor boundary (using continuous gradient magnitude and texture features) and shape features, as compared to the same features extracted from the entire tumor. In a 3-fold validation framework, gradient and texture features at tumor boundary, along with tumor shape, better predicted recurrence (mean AUC of 0.78 (std 0.15)) as compared to the same features from the entire tumor (mean AUC of 0.35 (std 0.11)). The addition of clinical variables did not improve the AUC. Using clinical variables alone, AUC of 0.6 (std 0.09) was obtained. Using BN LLM an example patient specific report in natural language is summarized below: Patient A clinical profile : Tumor size – large; age – younger (&amp;lt;50 y); KI67 – high; lymph node invasion – yes; peritumoral probability – high; grade – high. Probability of recurrence based on tumor boundary radiomics: High Top ranking clinical parameter: Younger age (&amp;lt;50 y) has high probability of recurrence. Prediction: Patient A is at high risk for recurrence. Conclusions: Heterogeneity features at the tumor boundary quantified by continuous gradient magnitude and texture features in routine clinical mammograms could predict TNBC recurrence. To the best of our knowledge, this is the first time that BN LLM has been used to generate patient specific clinically explainable report of risk for TNBC patients. This model will be validated in a larger cohort with a holdout validation set in future studies. Citation Format: S. Ghose, S. Cho, C. Davis, S. Gandhi, L. Lan, A. Mansuri, H. Trivedi, Y. Polar, F. Ginty, S. Badve. Tumor Boundary and Shape Features are Predictive of Recurrence in Triple Negative Breast Cancer (TNBC) Patients [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PD6-04.

BACKGROUND Accurate classification of tumor grade is necessary for understanding tumor development critical in patient management. Radiomic features are gaining popularity in classifying the tumors with the application of various classifiers. We evaluate five classifiers using gray-level co-occurrence matrix (GLCM) features to classify low-grade gliomas (LGG) and high-grade gliomas (HGG). METHODS We included high resolution multi-modal MR images from pre-operative BraTS 2019 database. The database contains a total of 335 multi-modal MR images (259 HGG and 76 LGG) with manually-corrected segmentations of tumor compartments. Sixty four three-dimensional (3D) voxel-by-voxel GLCM feature images on T2-wighted (T2), fluid attenuated inversion recovery (FLAIR), and T1-weighted (T1) pre- and post-contrast images are computed. A total of 192 features within regions of enhanced tumor (ET), edema (ED), and non-enhanced tumor (NET) are obtained by taking averages of the GLCM features within each region. For classification, we evaluated k-nearest neighbor (kNN), learning vector quantization (LVQ), random forest classifier (RF), classification and regression trees (CART), and support vector machine (SVM) classifiers for differentiating LGG from HGG. The dataset is randomly split into ratio of 80 to 20 for training and validation. The models are trained using repeated 5-fold cross-validation. Best 10 models for each of the classifiers are selected based on accuracy by applying multiple random split. RESULTS The average accuracies of 10 best models selected for each of kNN, LVQ, RF, CART, and SVM classifiers are 0.88, 0.92, 1.00, 0.890, and 0.95 on training set, and 0.89, 0.88, 0.88, 0.89, and 0.90 on validation set. The performance of five classifiers on validation set is similar. The accuracy of SVM classifier is slightly higher on validation set even though the RF appears to be the best classifier on training set. CONCLUSION Voxel-by-voxel GLCM features help differentiate LGG and HGG with 0.89 of accuracy irrespective of the classifier.

Ultrasound (US) radiomics analysis is an emerging research field to overcome clinician’s subjectivity of visual image assessment and interpretation. However, its clinical utility is still limited and the efficacy depends on the robustness of radiomic features. The purpose of this work is to evaluate the robustness of US radiomic features with various scanning settings, including central frequency, focal length, and overall Brightness Gain (BG). We tested the concept with Grey Level Co-occurrence Matrix (GLCM) features. All US images were acquired using a Hitachi Noblus US system and a bi-plane probe (EUP-U533C). The study utilized three materials: a tissue-mimicking phantom, beef muscle, and chicken breast. A total of 21 GLCM features were extracted from the US images. The relative percentage change was calculated as the standard deviation (STDEV) or the maximum difference (max difference) divided by the absolute mean value of each GLCM feature, varying in the BG = 19-29 zone. Among the 21 extracted GLCM features, we found seven robust features, namely differenceEntropy, entropy, homogeneity1, IDMN, IDN, inverseVariance, and sumEntropy, enduring within 10% variances when varying frequency, focal length, and BG settings. The results of this study indicate that some US radiomic features may be affected by scanning parameters, while others are more robust to these variations. As radiomics is expected to be a critical component for the integration of image-derived information to personalize treatment in the future, the robust features should be carefully chosen to obtain reliable radiomics-based analysis.