Breast Lesion Detection Using Weakly Dependent Customized Features and Machine Learning Models with Explainable Artificial Intelligence

Abstract

This research proposes a novel strategy for accurate breast lesion classification that combines explainable artificial intelligence (XAI), machine learning (ML) classifiers, and customized weakly dependent features from ultrasound (BU) images. Two new weakly dependent feature classes are proposed to improve the diagnostic accuracy and diversify the training data. These are based on image intensity variations and the area of bounded partitions and provide complementary rather than overlapping information. ML classifiers such as Random Forest (RF), Extreme Gradient Boosting (XGB), Gradient Boosting Classifiers (GBC), and LASSO regression were trained with both customized feature classes. To validate the reliability of our study and the results obtained, we conducted a statistical analysis using the McNemar test. Later, an XAI model was combined with ML to tackle the influence of certain features, the constraints of feature selection, and the interpretability capabilities across various ML models. LIME (Local Interpretable Model-Agnostic Explanations) and SHAP (SHapley Additive exPlanations) models were used in the XAI process to enhance the transparency and interpretation in clinical decision-making. The results revealed common relevant features for the malignant class, consistently identified by all of the classifiers, and for the benign class. However, we observed variations in the feature importance rankings across the different classifiers. Furthermore, our study demonstrates that the correlation between dependent features does not impact explainability.

Keywords: LIME; SHAP; XAI; dependent features; machine learning.

Figure 1

Bounded histogram features. (a) Raw breast US image from the BUSI dataset; (b) Ground truth of a breast lesion; (c) Region of interest. Selected pixels within bounded repartitions are shown according to their gray-levels distributions. (d1) [0, 31]; (d2) [32, 63]; (d3) [64, 95]; (d4) [96, 127]; (d5) [128, 159]; (d6) [160, 191]; (d7) [192, 223]; (d8) [224, 255].

Figure 2

The overall feature importance in the prediction results over the test dataset. (a) Bounded histogram features (Chi); (b) Grayscale density features (Ci). The most important features are marked in yellow.

Figure 3

LIME output: the importance of individual features in the classification process by their relevance and score, and the features’ selection across various classifiers. (a1) RF and CHi; (a2) RF and Ci; (b1) GBC and CHi; (b2) GBC and Ci; (c1) XGB and CHi; (c2) XGB and Ci. “0” or blue is associated with the malignant class and “1” or orange is for the benign class.

Figure 4

SHAP-integrated ML classifiers’ summary plot on the test data for the malignant and benign output classes. (a1) RF and CHi; (a2) RF and Ci; (b1) GBC and CHi; (b2) GBC and Ci; (c1) XGB and CHi; (c2) XGB and Ci. The horizontal axis plots an SV for a specific feature and data point. The vertical axis ranks the features based on their importance. The values of the features are represented with the following code: lower values are shown in blue, and higher values are shown in red. Points that overlap are shown vertically.