Fully automatic classification of breast MRI background parenchymal enhancement using a transfer learning approach

Abstract

Marked enhancement of the fibroglandular tissue on contrast-enhanced breast magnetic resonance imaging (MRI) may affect lesion detection and classification and is suggested to be associated with higher risk of developing breast cancer. The background parenchymal enhancement (BPE) is qualitatively classified according to the BI-RADS atlas into the categories “minimal,” “mild,” “moderate,” and “marked.” The purpose of this study was to train a deep convolutional neural network (dCNN) for standardized and automatic classification of BPE categories. This IRB-approved retrospective study included 11,769 single MR images from 149 patients. The MR images were derived from the subtraction between the first post-contrast volume and the native T1-weighted images. A hierarchic approach was implemented relying on 2 dCNN models for detection of MR-slices imaging breast tissue and for BPE classification, respectively. Data annotation was performed by 2 board-certified radiologists. The consensus of the 2 radiologists was chosen as reference for BPE classification. The clinical performances of the single readers and of the dCNN were statistically compared using the quadratic Cohen's kappa. Slices depicting the breast were classified with training, validation, and real-world (test) accuracies of 98%, 96%, and 97%, respectively. Over the 4 classes, the BPE classification was reached with mean accuracies of 74% for training, 75% for the validation, and 75% for the real word dataset. As compared to the reference, the inter-reader reliabilities for the radiologists were 0.780 (reader 1) and 0.679 (reader 2). On the other hand, the reliability for the dCNN model was 0.815. Automatic classification of BPE can be performed with high accuracy and support the standardization of tissue classification in MRI.

Breast MRI Background Parenchymal Enhancement Categorization Using Deep Learning: Outperforming the Radiologist

“…In prior work, BPE quantification methods included an automated FGT segmentation step and a calculation of percent enhancement, 13,14 but these methods show only slight to fair agreement with radiologists ( κ = 0.20–0.36) and are more discordant at higher levels of BPE, precisely when they might be more clinically useful 2 . More recently, supervised deep learning models have been used to approximate radiology report BPE designations, with test set accuracies ranging from 0.70 to 0.93 17,18,20,21 . In contrast to prior work, we show that an AI model not just approximates the radiology report BPE classification but can surpass it.…”

“…Nam et al developed a similar model using qualitative FGT and BPE scoring by radiologists as the ground truth. Most recently, Borkowski et al developed a fully automated BPE classification modal using sequential “breast slice detection” and a “BPE classification” neural networks, achieving noninferior performance to experienced radiologists 20 . However, no BPE tool to date has been shown to surpass current standard‐of‐care radiologist BPE designations, which is the standard clinical practice today.…”

Background Parenchymal Enhancement on Breast MRI: Assessment and Clinical Implications

“…Quantification of BPE volume and intensity on MRI may be feasible in the future, but we await publication of robust data on that topic before endorsing percentage recommendations'' [1]. Several different quantitative approaches have been investigated since the publication of the fifth edition of the BI-RADS Atlas, with numerous studies in the past five years [21][22][23][24][25][26][27][28][29][30][31][32][33][34].…”

“…Various semi-automated and automated segmentation models have been created [27][28][29][30]. BPE assessment tools using machine learning have also been developed [31][32][33][34]. Overall, many of these approaches have correlated well with qualitative BPE assessment.…”

A review of transfer learning for medical image classification

“…Feature extractor hybrid [56] Fine-tuning scratch [17,57] Many [23,25,[58][59][60][61][62][63] Microscopy Pathology Feature extractor [64][65][66][67][68][69][70] Fine-tuning [21] Fine-tuning scratch [71,72] MRI Bones Many [73] Genital systems Feature extractor [14,74] Integumentary system Fine-tuning scratch [75] Many [76] Nervous system Fine-tuning scratch [16,77,78] Many [19,[79][80][81] OCT Integumentary system Feature extractor [82] Cardiovascular system Many [83] Sense organs Feature extractor [84][85][86][87] Feature extractor hybrid [88] Fine-tuning [20] Fine-tuning scratch [29,89] Many [90][91][92] Photography Integumentary system Feature extractor…”