Speaker
Description
Distinguishing malignant from benign lesions using MRI imaging can significantly improve both diagnostic and therapeutic management. Artificial Intelligence (AI) can play a crucial role in this process. Dynamic Contrast-Enhanced (DCE) MRI has proven effective at this scope by providing valuable features related to cellularity and neoangiogenesis. We propose an ensemble learning approach to classify breast lesions leveraging morphological and dynamic features.
The analysis has been carried out on the publicly available “Advanced MRI Breast Lesions” dataset from The Cancer Imaging Archive. This dataset includes T2-weighted and DCE-MRI sequences (5 time-step), along with segmentations of all suspicious lesions. The 164 available lesions were divided into a training set (144 lesions) and a test set (20 lesions), with the latter taken apart to assess the generalization capability of the model. After extracting radiomic features using Pyradiomics python package, we computed dynamic features from DCE-MRI kinetic curves, which describe the contrast wash-in and wash-out. These features have been defined as the derivatives of measures, like mean or standard deviation, computed on the 5 DCE-MRI time steps inside the lesion masks. We trained and evaluated an XGBoost model, experimenting with different feature combinations in a stratified 5-fold cross-validation.
The best model trained on T2-weighted MRI morphological features achieved an Area Under the Curve (AUC) score of 0.83±0.02 on the independent test set, while the model using only dynamic features performed an AUC of 0.91±0.02.
Preliminary results show the great potential of integrating information present in DCE images. In the future, we will apply the analysis in a multicentric setting.