Deep learning of quantitative ultrasound multi-parametric images at pre-treatment to predict breast cancer response to chemotherapy

Abstract

In this study, a novel deep learning-based methodology was investigated to predict breast cancer response to neo-adjuvant chemotherapy (NAC) using the quantitative ultrasound (QUS) multi-parametric imaging at pre-treatment. QUS multi-parametric images of breast tumors were generated using the data acquired from 181 patients diagnosed with locally advanced breast cancer and planned for NAC followed by surgery. The ground truth response to NAC was identified for each patient after the surgery using the standard clinical and pathological criteria. Two deep convolutional neural network (DCNN) architectures including the residual network and residual attention network (RAN) were explored for extracting optimal feature maps from the parametric images, with a fully connected network for response prediction. In different experiments, the features maps were derived from the tumor core only, as well as the core and its margin. Evaluation results on an independent test set demonstrate that the developed model with the RAN architecture to extract feature maps from the expanded parametric images of the tumor core and margin had the best performance in response prediction with an accuracy of 88% and an area under the receiver operating characteristic curve of 0.86. Ten-year survival analyses indicate statistically significant differences between the survival of the responders and non-responders identified based on the model prediction at pre-treatment and the standard criteria at post-treatment. The results of this study demonstrate the promising capability of DCNNs with attention mechanisms in predicting breast cancer response to NAC prior to the start of treatment using QUS multi-parametric images.

Improving breast cancer diagnosis by incorporating raw ultrasound parameters into machine learning

“…However, this study only utilized one parameter for the comparison, while further improvements in breast lesion identification can utilize multiparametric analysis with deep learning approaches or machine learning classifiers. For example, a recent study (Taleghamar et al 2022) attempted a deep learning approach to predict breast treatment response using quantitative ultrasound (QUS) multiparametric maps, yielding results comparable with previous deep learning studies. Taleghamar et al extracted features from raw ultrasound signals (RF data) and added this information as 'preprocessing' .…”

“…These findings outperform previous research (Jarosik et al 2020, Byra et al 2022, Gare et al 2022 on utilizing RF data as AI input for breast classification, in which AUC were between 0.77 and 0.92. Moreover, there are studies (Uniyal et al 2014, Taleghamar et al 2022 utilizing RF data and extracting features for machine learning inputs with a similar approach as this study. Uniyal et al extracted features from RF ensembles or patches, used SVM for breast mass classification, and generated a malignancy probability map, of which AUC is 0.86; thus, our features improved breast lesion classification.…”

Implementation of Non-Invasive Quantitative Ultrasound in Clinical Cancer Imaging

“…Tissue microstructures and their scattering properties are distinct between invasive and non-invasive tumors [ 16 ], and from responsive to treatment compared to those of non-responsive ones [ 17 ]. The QUS spectral parametric images can provide a surrogate delineation of tumor microstructures providing diagnostic [ 18 , 19 ] and prognostic potentials [ 20 , 21 , 22 , 23 , 24 , 25 ]. QUS spectral parametric maps are created using a sliding window technique.…”

“…Along with the corresponding attenuation functions and reference backscatter coefficient (BSC), this allows for the estimation of the tissue-dependent scattering function. Parametrization of the frequency-dependent scattering function results in the linear-fit and acoustic scattering parameters useful for diagnostics and prognostics [ 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 ].…”

“…From the parametric maps, radiomics features can be extracted that include first-order statistical, morphological, and textural features. These radiomic features are potential imaging biomarkers for diagnosis and prognosis purposes [ 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 ]. Radiomics texture features rely on the hypothesis that tissue heterogeneity can be quantified through their surrogate spectral parametric maps ( Figure 1 ).…”