Machine Learning Models and Multiparametric Magnetic Resonance Imaging for the Prediction of Pathologic Response to Neoadjuvant Chemotherapy in Breast Cancer

Abstract

Background: Most breast cancer (BC) patients fail to achieve pathological complete response (pCR) after neoadjuvant chemotherapy (NAC). The aim of this study was to evaluate whether imaging features (perfusion/diffusion imaging biomarkers + radiomic features) extracted from pre-treatment multiparametric (mp)MRIs were able to predict, alone or in combination with clinical data, pCR to NAC. Methods: Patients with stage II-III BC receiving NAC and undergoing breast mpMRI were retrospectively evaluated. Imaging features were extracted from mpMRIs performed before NAC. Three different machine learning models based on imaging features, clinical data or imaging features + clinical data were trained to predict pCR. Confusion matrices and performance metrics were obtained to assess model performance. Statistical analyses were conducted to evaluate differences between responders and non-responders. Results: Fifty-eight patients (median [range] age, 52 [45–58] years) were included, of whom 12 showed pCR. The combined model improved pCR prediction compared to clinical and imaging models, yielding 91.5% of accuracy with no false positive cases and only 17% false negative results. Changes in different parameters between responders and non-responders suggested a possible increase in vascularity and reduced tumour heterogeneity in patients with pCR, with the percentile 25th of time-to-peak (TTP), a classical perfusion parameter, being able to discriminate both groups in a 75% of the cases. Conclusions: A combination of mpMRI-derived imaging features and clinical variables was able to successfully predict pCR to NAC. Specific patient profiles according to tumour vascularity and heterogeneity might explain pCR differences, where TTP could emerge as a putative surrogate marker for pCR.

Accessible Machine Learning and Deep Learning Models Predict Response and Survival in Early Stage Hormone Receptor-Positive/HER2-Negative Breast Cancer Receiving Neoadjuvant Chemotherapy

“…A high recurrence risk score of gene expression panels such as the 21-gene assay on diagnostic biopsy has been shown to predict pCR 47,48 , as well as the assessment of circulating tumor DNA and particularly its early dynamic changes during over the course of NCT 49 . Furthermore, in the BC NCT setting, ML and deep learning models have been explored to predict the pathological response based on pre-NCT histopathological and radiological images [50][51][52] . Thus, the model we speci cally developed for HoR-positive/HER2-negative subtype could be integrated with digital pathology images and radiology features.…”

Radiomics of Automated Breast Ultrasound for the Pretreatment Prediction of the Pathological Response to Neoadjuvant Chemotherapy in Breast Cancer

“…The study conducted by Kun Chen et al [9] established four PET/CT-based radiomics models that effectively predict pCR in breast cancer patients undergoing NAC, and the AUC of these models ranged from 0.70 to 0.77. Radiomics of multiparametric MRI images can also be employed for predicting the efficacy before NAC [27][28][29][30] . Zhenyu Liu et al [6] developed a radiomics model based on multiparametric MRI images before NAC, with an AUC of 0.79.…”

Impact of Imaging Biomarkers and AI on Breast Cancer Management: A Brief Review

“…Their study demonstrated that integrating radiomics features and imaging features extracted from DWI-MRI and DCE-MRI, such as gray-level size-zone matrix, GLCM, and gray-level run length matrix, along with clinical data such as molecular subtype and clinical tumor stage, significantly enhanced performance compared to using clinical or imaging features alone, achieving an accuracy of 91.5%. Collectively, the literature studies presented [233,235,236,[238][239][240] underscore the potential for improving the pre-treatment prediction of tumor response to NAC through the application of AI/ML techniques.…”

“…The combination of radiomics features with molecular subtypes notably improved prediction performance, elevating the AUC from 0.72 to 0.80. Vicent et al [240] harnessed six machine learning algorithms (KNN, DT, RF, AdaBoost, GBoost, GNB, LDA, LR, and MLP) to predict tumor response to NAC, classifying it into two categories: pCR and no-pCR. They assessed their algorithms using data from 58 patients collected at Castellón provincial hospital, Spain.…”

“…• ML/AI components are able to provide quantifiable, objective measures for BC detection and diagnosis and can help with pre-treatment tumor response prediction to NAC. Therefore, their findings have the potential to enhance the effectiveness of the healthcare systems for BC [233,235,236,[238][239][240].…”