Radiomics-based machine learning analysis and characterization of breast lesions with multiparametric...

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Accurate characterization of breast lesions is crucial for timely diagnosis and treatment.
• Multiparametric diffusion-weighted imaging (DWI) offers advanced insights into tissue microstructure.
• Radiomics analysis extracts quantitative features from medical images, potentially enhancing diagnostic accuracy.

Purpose of the Study:

• To evaluate the utility of radiomics-based machine learning analysis with multiparametric DWI.
• To compare the diagnostic performance of radiomics features against mean diffusion metrics for breast lesion characterization.

Main Methods:

• A retrospective study of 542 breast lesions using mono-exponential (ME), biexponential (BE), stretched exponential (SE), and diffusion-kurtosis imaging (DKI).
• Computation of 100 radiomics features and comparison of four classifiers (random forest, PCA, L1R, SVM) via cross-validation.
• Calculation and comparison of diagnostic performance using area under the receiver operating characteristic curve (AUC) for radiomics features and mean diffusion metrics.

Main Results:

• Random forest (RF) achieved higher AUCs than other classifiers.
• Radiomics features demonstrated superior diagnostic performance (AUCs 0.80-0.85) compared to mean diffusion metrics (AUCs 0.54-0.79).
• Significant differences (P < 0.001) were observed between radiomics features and mean diffusion metrics for differentiating benign and malignant lesions, with BE_D (AUC: 0.85) being the most effective sequence.

Conclusions:

• Radiomics-based analysis of multiparametric DWI using RF provides superior differentiation of benign and malignant breast lesions.
• This approach offers a promising non-invasive tool for improving breast lesion characterization in clinical practice.