Characterization of breast lesions using multi-parametric diffusion MRI and machine learning

Abstract

Objective: To investigate quantitative imaging markers based on parameters from two diffusion-weighted imaging (DWI) models, continuous-time random-walk (CTRW) and intravoxel incoherent motion (IVIM) models, for characterizing malignant and benign breast lesions by using a machine learning algorithm. Approach: With IRB approval, 40 women with histologically confirmed breast lesions (16 benign, 24 malignant) underwent DWI with 11 b-values (50 to 3000 s/mm2) at 3T. Three CTRW parameters, Dm , α, and β and three IVIM parameters Ddiff , Dperf , and f were estimated from the lesions. A histogram was generated and histogram features of skewness, variance, mean, median, interquartile range; and the value of the 10%, 25%, and 75% quantiles were extracted for each parameter from the regions-of-interest. Feature significance was calculated using the Boruta algorithm using Benjamin Hochberg False Discover Rate and Bonferroni correction for hypothesis testing. Predictive performance of the significant features was evaluated using Support Vector Machine, Random Forest, Naïve Bayes, Gradient Boosted Classifier (GB), Decision Trees, AdaBoost and Gaussian Process machine learning classifiers. Main Results: The 75% quantile, and median of Dm ; 75% quantile of f; mean, median, and skewness of β; kurtosis of Dperf ; and 75% quantile of Ddiff were the most significant features. The GB differentiated malignant and benign lesions with an accuracy of 0.833, an area-under-the-curve of 0.942, and an F1 score of 0.87 providing the best statistical performance (p-value < 0.05) compared to the other classifiers. Significance: Our study has demonstrated that GB with a set of histogram features from the CTRW and IVIM model parameters can effectively differentiate malignant and benign breast lesions.

Ensemble Learning for Breast Cancer Lesion Classification: A Pilot Validation Using Correlated Spectroscopic Imaging and Diffusion-Weighted Imaging

“…This study showed the feasibility of using metabolite ratios from 5D EP-COSI and ADC values from the DWI data of breast cancer patients to train machine learning models for classifying benign and malignant lesions. While earlier studies have attempted lesion characterization using features extracted from the DWI and DCE-MRI data, these models did not use the quantitative measures of metabolite and lipid features which can be obtained with an MRSI examination [ 48 , 49 , 50 ]. Although variations in water and fat levels can become ambiguous in glandular regions, especially in benign and healthy tissues, various lipid and metabolite ratios are reported to have statistically significant differences between benign and malignant lesions [ 42 ].…”

“…Since the focus of this study was to analyze the performance of ML models with features from the 5D EP-COSI data, we have not considered some of the image-based features potentially available from DWI. For example, it has been recently shown that the features based on continuous-time random-walk (CTRW) and intravoxel incoherent motion (IVIM) models from DWI using multiple b-values can classify benign and malignant breast lesions using ensemble ML models [ 48 ]. More radiomics features from DWI as well as other modalities like DCE-MRI can be used in a future study to potentially further improve the model performance.…”

“…While individual machine learning models may perform well, a meta-approach that combines individual models named ensemble learning could generate even more generalizable models that can reduce individual base learner’s variance or bias [ 46 ]. In particular, advanced ensemble models like the gradient-boosted tree-based algorithm that combines multiple weak learners (decision trees) are shown to be capable of detecting key features of the multi-modal, multi-parametric imaging information for applications such as tissue/cancer grade classification [ 47 , 48 , 49 , 50 ].…”

“…Multiple studies have recently shown that the features extracted from DCE-MRI and DWI of breast tissues used in ML models are capable of predicting tumor grades and classifying benign and malignant breast lesions [ 48 , 49 , 50 ]. However, metabolite and lipid information from MRSI data has not been used in this context so far.…”

Classification of MRI-only suspicious breast lesions: Development and validation of a nomogram combining MRI features and apparent diffusion coefficient histogram

“…Diffusion-weighted imaging (DWI) is a fast, noninvasive, and non-enhancing functional imaging method and is an integral part of the clinical breast MRI protocol (10). The apparent diffusion coe cient (ADC) map derived from DWI can quantitatively measure the diffusion hindrance of water molecules in the lesion tissue and provide quantitative parameters for lesion characterization (11)(12)(13). Histogram analysis can evaluate the distribution and variation of intensity of all voxels in the lesion by extracting more objective quantitative data from medical images and thus provide richer information for lesion characterization, which has the advantages of simplicity and good reproducibility (14)(15)(16).…”