Early prediction of pathological complete response to neoadjuvant chemotherapy in breast cancer MRI images using combined Pre-trained convolutional neural network and machine learning. (15th February 2023)

• Record Type: Journal Article
• Authors: Khanna, Priyanka; Sahu, Mridu; Kumar Singh, Bikesh; Bhateja, Vikrant
• Is Part Of: Measurement. Volume 207(2023)
• Publication Date: 2023-02-15
• Subjects: Breast cancer -- Dynamic contrast-enhanced MRI -- Neoadjuvant chemotherapy -- Machine learning -- Deep learning; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.112269
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A hybrid approach for predicting Chemotherapy response is proposed. Breast cancer MRI images(first visit) are used for evaluation. Customized pretrained Convolution neural network is used for feature extraction. Extensive comparative evaluation is conducted using machine learning techniques. A higher classification accuracy is achieved using the hybrid approach. Abstract: Objective: Timely diagnosis of breast cancer can ameliorate the treatment plan, thus reducing the mortality rate. We propose a model integrating pre-trained Convolutional neural network (CNN) with machine learning for prognosticating pathologic complete response(PCR) using breast cancer dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) prior to commencement of neoadjuvant chemotherapy(NACT). For predicting pathologic complete response (PCR) using dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) for breast cancer prior to the start of neoadjuvant chemotherapy, we present a hybrid model integrating a pre-trained Convolutional neural network (CNN) with machine learning (NACT). Material & Methods: In this retrospective study, 64 patients receiving NACT for invasive breast cancer are examined. Deep learning-based pre-trained CNN models ResNet-50 and ResNet-18 were used to extract features from patient visit 1 MRI images (before the initiation of NACT). Mann-Whitney U tests is used to assess features and their relevance (significance level p less than 0.05 and confidence interval is 95 %). Furthermore, features extracted and features selected were independently given as an input to different machine learning classifiers for the prediction of response of NACT. Classification performance was assessed under different data division protocols using accuracy, specificity, sensitivity, and area under the receiver operating characteristic curve (AUROC). Result: The hybrid combination using ResNet-18 as feature extractor, fine K-nearest neighbor(KNN) as classifier and feature selected using Mann-Whitney U test outperformed the result. Accuracy of 99.8% and AUROC of 1 is obtained under hold-out validation protocol while accuracy of 99.3%, and AUROC of 0.99 is obtained under 10-fold cross-validation. Conclusion: The proposed model employing DCE-MRI images acquired before starting chemotherapy has considerable accuracy in classifying PCR and non-PCR patients. The efficacy of the prediction model can improve considerably on the back of a larger dataset.