Using an Improved Residual Network to Identify PIK3CA Mutation Status in Breast Cancer on Ultrasound Image

Abstract

BackgroundThe detection of phosphatidylinositol-3 kinase catalytic alpha (PIK3CA) gene mutations in breast cancer is a key step to design personalizing an optimal treatment strategy. Traditional genetic testing methods are invasive and time-consuming. It is urgent to find a non-invasive method to estimate the PIK3CA mutation status. Ultrasound (US), one of the most common methods for breast cancer screening, has the advantages of being non-invasive, fast imaging, and inexpensive. In this study, we propose to develop a deep convolutional neural network (DCNN) to identify PIK3CA mutations in breast cancer based on US images.Materials and MethodsWe retrospectively collected 312 patients with pathologically confirmed breast cancer who underwent genetic testing. All US images (n=800) of breast cancer patients were collected and divided into the training set (n=600) and test set (n=200). A DCNN-Improved Residual Network (ImResNet) was designed to identify the PIK3CA mutations. We also compared the ImResNet model with the original ResNet50 model, classical machine learning models, and other deep learning models.ResultsThe proposed ImResNet model has the ability to identify PIK3CA mutations in breast cancer based on US images. Notably, our ImResNet model outperforms the original ResNet50, DenseNet201, Xception, MobileNetv2, and two machine learning models (SVM and KNN), with an average area under the curve (AUC) of 0.775. Moreover, the overall accuracy, average precision, recall rate, and F1-score of the ImResNet model achieved 74.50%, 74.17%, 73.35%, and 73.76%, respectively. All of these measures were significantly higher than other models.ConclusionThe ImResNet model gives an encouraging performance in predicting PIK3CA mutations based on breast US images, providing a new method for noninvasive gene prediction. In addition, this model could provide the basis for clinical adjustments and precision treatment.

A nomogram model combining ultrasound-based radiomics features and clinicopathological factors to identify germline BRCA1/2 mutation in invasive breast cancer patients

“…However, similar studies that included gray-scale images alone yielded limited AUCs of 0.651–0.685 [ 40 , 41 ]. For breast cancer, in a study utilizing gray-scale US images of 312 cases to predict PIK3CA mutation, several machine learning and deep learning models were constructed and showed relatively strong predictive power (AUC 0.741–0.775) [ 42 ].…”

Effect of caspase-1 (CASP1) combined with multimodal ultrasound features on the prognosis of breast cancer patients

“…At the same time, we also performed bioinformatics analysis and verification to determine the effect of CASP1 on tumor cell behavior. Then, we evaluated CASP1 in tandem with the multimodal ultrasonography characteristics of study participants with BRCA to further assess its prognostic efficacy ( 33 , 34 ).…”

Computational pathology to improve biomarker testing in breast cancer: how close are we?

“…This approach takes the raw medical images as inputs, does not require manually designed features, and can automatically learn features related to classification or segmentation tasks (LeCun et al , 2015). The above study observes the differences in breast morphology and other features resulting from microstructural changes in PIK3CA mutant breast cancers and investigated whether the differences could be captured and interpreted by ultrasound images (Shen et al ., 2022).…”

“…Although these methods for detecting genetic mutations have improved considerably, molecular testing is often time-consuming, operator dependent, and may be limited by inadequate sample availability. For these reasons, developing an artificial intelligence-based approach to test these biomarkers' alterations could overcome many challenges faced by traditional molecular testing techniques (Shen et al, 2022). For instance, an attempt to overcome traditional testing methods for PIK3CA mutational status assessment was accomplished by Shen et al (Shen et al, 2022), with the development of a deep convolutional neural network (DCNN) to identify PIK3CA mutations in breast cancer based on ultrasound images.…”

“…For these reasons, developing an artificial intelligence-based approach to test these biomarkers' alterations could overcome many challenges faced by traditional molecular testing techniques (Shen et al, 2022). For instance, an attempt to overcome traditional testing methods for PIK3CA mutational status assessment was accomplished by Shen et al (Shen et al, 2022), with the development of a deep convolutional neural network (DCNN) to identify PIK3CA mutations in breast cancer based on ultrasound images. Unlike traditional machine learning and radiomics methods, DCNN, a special type of deep learning, does not require domain experts to select the specific features beforehand.…”