Radiomics and machine learning are rapidly growing fields of breast MRI research. We are investigating the use of these techniques in the prediction of pathological complete response to neoadjuvant chemotherapy (NACT). NACT is used in the management of breast cancer to reduce the size of tumours before surgery, improving the rate of breast conservation surgery and reducing the extent of axillary surgery. The best outcome from NACT is pathological complete response to chemotherapy (pCR), defined as the absence of residual invasive carcinoma in the breast or lymph nodes. MRI is the most accurate technique for the assessment of response to treatment, and patients are generally imaged before, during, and after their course of chemotherapy. Assessment of response during treatment is crucial to avoid unnecessary toxicity and cost by stopping treatment or changing to a more effective regime if a patient is not showing therapeutic response or if a patient has achieved pCR. Early prediction of response at the start of chemotherapy could provide a more personalised approach to treatment, allowing for better timing of surgery, choice of treatment regimen and a more individualised prognosis.
In this work, large numbers of quantitative radiomics features are extracted from images and are used to train machine learning models for classification. Combining the kinetic, textural, and morphological features derived from MRI, as well as clinicopathological features, using machine learning approaches can provide a more sophisticated method of predicting pCR and better our understanding of imaging biomarkers of tumour aggressiveness. In contrast to use of radiomics features, convolutional neural networks (CNNs) are also being investigated for the prediction of pCR, eliminating the need for the choice of features and the time-consuming manual delineation of breast lesions by breast radiologists.