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.

Purpose: To develop deep learning models for predicting the pathologic complete response (pCR) to neoadjuvant systemic therapy (NAST) in patients with triple-negative breast cancer (TNBC) based on pretreatment multiparametric breast MRI and clinicopathological data. Methods: The prospective institutional review board-approved study [NCT02276443] included 282 patients with stage I–III TNBC who had multiparametric breast MRI at baseline and underwent NAST and surgery during 2016–2021. Dynamic contrast-enhanced MRI (DCE), diffusion-weighted imaging (DWI), and clinicopathological data were used for the model development and internal testing. Data from the I-SPY 2 trial (2010–2016) were used for external testing. Four variables with a potential impact on model performance were systematically investigated: 3D model frameworks, tumor volume preprocessing, tumor ROI selection, and data inputs. Results: Forty-eight models with different variable combinations were investigated. The best-performing model in the internal testing dataset used DCE, DWI, and clinicopathological data with the originally contoured tumor volume, the tight bounding box of the tumor mask, and ResNeXt50, and achieved an area under the receiver operating characteristic curve (AUC) of 0.76 (95% CI: 0.60–0.88). The best-performing models in the external testing dataset achieved an AUC of 0.72 (95% CI: 0.57–0.84) using only DCE images (originally contoured tumor volume, enlarged bounding box of tumor mask, and ResNeXt50) and an AUC of 0.72 (95% CI: 0.56–0.86) using only DWI images (originally contoured tumor volume, enlarged bounding box of tumor mask, and ResNet18). Conclusions: We developed 3D deep learning models based on pretreatment data that could predict pCR to NAST in TNBC patients.

Accurate tumor segmentation is required for quantitative image analyses, which are increasingly used for evaluation of tumors. We developed a fully automated and high-performance segmentation model of triple-negative breast cancer using a self-configurable deep learning framework and a large set of dynamic contrast-enhanced MRI images acquired serially over the patients’ treatment course. Among all models, the top-performing one that was trained with the images across different time points of a treatment course yielded a Dice similarity coefficient of 93% and a sensitivity of 96% on baseline images. The top-performing model also produced accurate tumor size measurements, which is valuable for practical clinical applications.

We have developed a family of biology-based mathematical models of high-grade glioma (HGG), capturing the key features of tumour growth and response to chemoradiation. We now seek to quantify the accuracy of parameter estimation and determine, when given a virtual patient cohort, which model was used to generate the tumours. In this way, we systematically test both the parameter and model identifiability. Virtual patients are generated from unique growth parameters whose growth dynamics are determined by the model family. We then assessed the ability to recover model parameters and select the model used to generate the tumour. We then evaluated the accuracy of predictions using the selected model at four weeks post-chemoradiation. We observed median parameter errors from 0.04% to 72.96%. Our model selection framework selected the model that was used to generate the data in 82% of the cases. Finally, we predicted the growth of the virtual tumours using the selected model resulting in low error at the voxel-level (concordance correlation coefficient (CCC) ranged from 0.66 to 0.99) and global level (percentage error in total tumour cellularity ranged from −12.35% to 0.07%). These results demonstrate the reliability of our framework to identify the most appropriate model under noisy conditions expected in the clinical setting. This article is part of the theme issue 'Uncertainty quantification for healthcare and biological systems (Part 2)'.