Improving Breast Mass Classification Performance of Radiomics-Based Model by Image Enhancement with Discrete Wavelet Transformation

Early detection and accurate characterization of breast mass play an essential role in breast cancer treatment, one of the leading risks to women. Taking advantage of the development of artificial intelligence-based computer-aid diagnosis tools, the Radiomics-based machine learning model presented a promising performance in breast mass classification by extracting massive amounts of robust features within the abnormal region. In this work, the utilization of Discrete Wavelet Transformation (DWT) was employed to boost the low-frequency patterns of the breast ultrasound image to enhance its performance. These low-frequency patterns potentially contain valuable information to distinguish benign and malignant breast masses. Firstly, the 1D DWT was applied to get enhanced images, and then a total of 80 Radiomics-based features were extracted from the enhanced image dataset. In the training stage, three Machine Learning models, namely Support Vector Machine, Random Forest, and XGBoost, were utilized. Finally, the proposed pipeline (DWT-Radiomics) classification performance was compared to the conventional Radiomics pipeline by the 4-fold cross-validation technique. The DWT-Radiomics evaluation metrics are the most important in enhanced images. Especially in the SVM model, the weighted F1, Precision, and recall are 0.675, 0.715, and 0.703 respectively, compared to 0.553, 0.465, and 0.682 respectively in the conventional pipeline. The XGBoost model achieved the highest performance with the weighted F1, precision, and recall scores were 0.800, 0.801, and 0.802, respectively in the DWT-Radiomics pipeline and 0.774, 0.773, and 0.777 in the conventional pipeline. Moreover, the mutual information index of DWT-based features is significantly greater than the conventional feature. The results present that DWT-Radomics feature extraction out-performed conventional Radiomics in benign and malignant mass discrimination and model classification. In conclusion, the DWT could enhance robust patterns that significantly contribute to breast mass classification.