Prediction of Malignant Breast Lesions from MRI Features: A Comparison of Artificial Neural Network and Logistic Regression Techniques

doi:10.1016/j.acra.2009.01.029

Academic Radiology

Volume 16, Issue 7, July 2009, Pages 842-851

https://doi.org/10.1016/j.acra.2009.01.029 Get rights and content

Rationale and Objectives

Dynamic contrast-enhanced magnetic resonance imaging is a clinical imaging modality for the detection and diagnosis of breast lesions. Analytic methods were compared for diagnostic feature selection and the performance of lesion classification to differentiate between malignant and benign lesions in patients.

Materials and Methods

The study included 43 malignant and 28 benign histologically proved lesions. Eight morphologic parameters, 10 gray-level co-occurrence matrix texture features, and 14 Laws texture features were obtained using automated lesion segmentation and quantitative feature extraction. Artificial neural network (ANN) and logistic regression analysis were compared for the selection of the best predictors of malignant lesions among the normalized features.

Results

Using ANN, the final four selected features were compactness, energy, homogeneity, and Law_LS, with an area under the receiver-operating characteristic curve (AUC) of 0.82 and accuracy of 0.76. The diagnostic performance of these four features computed on the basis of logistic regression yielded an AUC of 0.80 (95% confidence interval [CI], 0.688–0.905), similar to that of ANN. The analysis also showed that the odds of a malignant lesion decreased by 48% (95% CI, 25%–92%) for every increase of 1 standard deviation in the Law_LS feature, adjusted for differences in compactness, energy, and homogeneity. Using logistic regression with z-score transformation, a model composed of compactness, normalized radial length entropy, and gray-level sum average was selected, and it had the highest overall accuracy, 0.75, among all models, with an AUC of 0.77 (95% CI, 0.660–0.880). When logistic modeling of transformations using the Box-Cox method was performed, the most parsimonious model with predictors compactness and Law_LS had an AUC of 0.79 (95% CI, 0.672–0.898).

Conclusion

The diagnostic performance of models selected by ANN and logistic regression was similar. The analytic methods were found to be roughly equivalent in terms of predictive ability when a small number of variables were chosen. The robust ANN methodology uses a sophisticated nonlinear model, while logistic regression analysis provides insightful information to enhance the interpretation of the model features.

Section snippets

Malignant and Benign Lesion Database

The database analyzed in this study included 43 malignant and 28 benign lesions, the same database as reported in a previous study (18). The 43 patients with malignant lesions were aged 29 to 76 years (mean ± standard deviation, 48 ± 9 years; median, 48 years). The 28 patients with benign lesions were aged 21 to 74 years (mean, 45 ± 7 years; median, 45 years). The MRI studies were performed using a Philips Eclipse 1.5-T scanner (Philips, Cleveland, OH). The images were acquired using a

ANN Diagnostic Feature Selection and Evaluation

Before ANN training, features were transformed to z scores on the basis of the mean and SD of the entire set of 71 lesions. A three-layer ANN with 5 hidden nodes in the hidden layer was chosen after a number of trial-and-error runs. The diagnostic measures used to assess differentiation between 43 malignant and 28 benign lesions are summarized in Table 1. Considering the eight morphology features, the classifier selected by ANN included three parameters: lesion volume, NRL entropy, and

Discussion

The primary aim of this study was to compare ANN and logistic regression analysis for lesion classification to differentiate between malignant and benign breast lesions in patients. Using our data set of 71 lesions, the ANN procedure was applied to select the best classifiers for morphologic and texture (GLCM and Laws) category features. The three selected morphologic features (volume, NRL entropy, and compactness) achieved a moderate AUC of 0.80 and estimated accuracy of 0.77. The three

References (40)

W. Chen et al.
A fuzzy c-means (FCM)-based approach for computerized segmentation of breast lesions in dynamic contrast-enhanced MR images
Acad Radiol
(2006)
K. Schelfout et al.
Contrast-enhanced MR imaging of breast lesions and effect on treatment
Eur J Surg Oncol
(2004)
Y.H. Chou et al.
Stepwise logistic regression analysis of tumor contour features for breast ultrasound diagnosis
Ultrasound Med Biol
(2001)
J.V. Tu
Advantages and disadvantages of using artificial neural networks versus logistic regression for predicting medical outcomes
J Clin Epidemiol
(1996)
J.H. Song et al.
Comparative analysis of logistic regression and artificial neural network for computer-aided diagnosis of beast masses
Acad Radiol
(2005)
K. Nie et al.
Quantitative analysis of lesion morphology and texture features for diagnostic prediction in breast MRI
Acad Radiol
(2008)
D. Delen et al.
Predicting breast cancer survivability: a comparison of three data mining methods
Artif Intell Med
(2005)
N. Terrin et al.
External validity of predictive models: a comparison of logistic regression, classification trees, and neural networks
J Clin Epidemiol
(2003)
W. Chen et al.
Automatic identification and classification of characteristic kinetic curves of breast lesions on DCE-MRI
Med Phys
(2006)
W. Chen et al.
Computerized interpretation of breast MRI: investigation of enhancement-variance dynamics
Med Phys
(2004)

G.P. Liney et al.

Breast lesion analysis of shape technique: semiautomated vs. manual morphological description

J Magn Reson Imaging

(2006)

L.A. Meinel et al.

Breast MRI lesion classification: improved performance of human readers with a backpropagation neural network computer-aided diagnosis (CAD) system

J Magn Reson Imaging

(2007)

L. Esserman et al.

Utility of magnetic resonance imaging in the management of breast cancer: evidence for improved preoperative staging

J Clin Oncol

(1999)

U. Fischer et al.

Breast carcinoma: effect of preoperative contrast-enhanced MR imaging on the therapeutic approach

Radiology

(1999)

H. Mumtaz et al.

Staging of symptomatic primary breast cancer with MR imaging

AJR Am J Roentgenol

(1997)

A. Rieber et al.

Pre-operative staging of invasive breast cancer with MR mammography and/or PET: boon or bunk?

Br J Radiol

(2002)

Y. Zhang et al.

The role of contrast-enhanced MR mammography for determining candidates for breast conservation surgery

Breast Cancer

(2002)

K.G. Gilhuijs et al.

Computerized analysis of breast lesions in three dimensions using dynamic magnetic-resonance imaging

Med Phys

(1998)

H.P. Chan et al.

Computerized classification of malignant and benign microcalcifications on mammograms: texture analysis using an artificial neural network

Phys Med Biol

(1997)

D.J. Sargent

Comparison of artificial neural networks with other statistical approaches

Cancer

(2001)

Cited by (75)

Performance of Machine Learning Methods Based on Multi-Sequence Textural Parameters Using Magnetic Resonance Imaging and Clinical Information to Differentiate Malignant and Benign Soft Tissue Tumors
2023, Academic Radiology
Citation Excerpt :
Image texture refers to a set of metrics for digital images calculated by mathematical analysis. Many studies have reported that textural analysis by imaging offers highly accurate diagnostic performance for tumors in the brain (10-12), kidney (13), liver (14), prostate (15,16), breast (17), and uterus (18). Many machine learning methods have been applied to image texture analysis (ITA) for diagnosing these tumors.
To evaluate the performance of a machine learning method to differentiate malignant from benign soft tissue tumors based on textural features on multiparametric magnetic resonance imaging (mpMRI).
We enrolled 163 patients with soft tissue tumors whose diagnosis was pathologically proven (71 malignant, 92 benign). All patients underwent mpMRI. Twelve histographic and textural parameters were assessed on T1-weighted imaging (T1WI), T2-weighted imaging, diffusion-weighted imaging, apparent diffusion coefficient maps, and contrast-enhanced T1WI imaging. We compared mean signals of all sequences from the malignant and benign tumors using Welch's t-test. Prediction models were developed via a machine learning technique (support vector machine) using textural features of each sequence, clinical information (sex + age + tumor size), and the combined model incorporating all features. Areas under the receiver operating characteristic curves (AUCs) of these models were calculated using fivefold cross validation.
The diagnostic ability of clinical information model (AUC 0.85) was not inferior to the model with textural features of each sequence (AUC 0.79–0.84). The combined model showed the highest diagnostic ability (AUC 0.89). The AUC of the combined model (0.89) was comparable to those of two board-certified radiologists (0.89 and 0.87).
Machine learning methods based on textural features on mpMRI and clinical information offer adequate diagnostic performance to differentiate between malignant and benign soft tissue tumors.
Magnetic resonance imaging histogram analysis of amygdala in functional neurological disorder: Histogram Analysis of Amygdala in Functional Neurological Disorder
2022, Psychiatry Research - Neuroimaging
Machine Learning to Differentiate T2-Weighted Hyperintense Uterine Leiomyomas from Uterine Sarcomas by Utilizing Multiparametric Magnetic Resonance Quantitative Imaging Features
2019, Academic Radiology
Citation Excerpt :
Many studies have reported that image texture analysis offers high diagnostic performance for tumors in the brain (7–9), soft tissue (10), kidney (11), liver (12), prostate (13,14), and breast (15), and a recent preliminary study suggested that texture analysis could also be used to differentiate uterine leiomyosarcoma from atypical leiomyoma (16). Various machine learning methods and algorithms have been used for these analyses, including logistic regression (14), support vector machines (7,10,17), multilayer perceptron artificial neural networks (10,15), and the k-nearest neighbors (8,12), naive Bayes (18), and random forest algorithms (9,13). Recently, a new machine learning method called eXtreme Gradient Boosting (XGBoost) was introduced by Chen and Guestrin (19), although it has not yet been used extensively for medical image analysis.
Uterine leiomyomas with high signal intensity on T2-weighted imaging (T2WI) can be difficult to distinguish from sarcomas. This study assessed the feasibility of using machine learning to differentiate uterine sarcomas from leiomyomas with high signal intensity on T2WI on multiparametric magnetic resonance imaging.
This retrospective study included 80 patients (50 with benign leiomyoma and 30 with uterine sarcoma) who underwent pelvic 3 T magnetic resonance imaging examination for the evaluation of uterine myometrial smooth muscle masses with high signal intensity on T2WI. We used six machine learning techniques to develop prediction models based on 12 texture parameters on T1WI and T2WI, apparent diffusion coefficient maps, and contrast-enhanced T1WI, as well as tumor size and age. We calculated the areas under the curve (AUCs) using receiver-operating characteristic analysis for each model by 10-fold cross-validation and compared these to those for two board-certified radiologists.
The eXtreme Gradient Boosting model gave the highest AUC (0.93), followed by the random forest, support vector machine, multilayer perceptron, k-nearest neighbors, and logistic regression models. Age was the most important factor for differentiation (leiomyoma 44.9 ± 11.1 years; sarcoma 58.9 ± 14.7 years; p < 0.001). The AUC for the eXtreme Gradient Boosting was significantly higher than those for both radiologists (0.93 vs 0.80 and 0.68, p = 0.03 and p < 0.001, respectively).
Machine learning outperformed experienced radiologists in the differentiation of uterine sarcomas from leiomyomas with high signal intensity on T2WI.
NCTN Assessment on Current Applications of Radiomics in Oncology
2019, International Journal of Radiation Oncology Biology Physics
Radiomics is a fast-growing research area based on converting standard-of-care imaging into quantitative minable data and building subsequent predictive models to personalize treatment. Radiomics has been proposed as a study objective in clinical trial concepts and a potential biomarker for stratifying patients across interventional treatment arms. In recognizing the growing importance of radiomics in oncology, a group of medical physicists and clinicians from NRG Oncology reviewed the current status of the field and identified critical issues, providing a general assessment and early recommendations for incorporation in oncology studies.
Decision support systems in breast cancer
2019, Precision Medicine for Investigators, Practitioners and Providers
Breast cancer is a highly heterogeneous disease, exhibiting substantial spatiotemporal variations in gene expression, biochemistry, histopathology, and macroscopic structure. The complexities in the evaluation of malignant tumor heterogeneity, are pushing current clinical examinations to their limits and accounting for most of the failures of targeted therapies, and uncertainties in the clinical outcome among breast cancer patients. The incorporation of state-of-the-art image analysis methods, and emerging machine learning techniques, is providing the application of detailed data-driven quantitative studies of imaging phenotypes, termed Radiomics. Specifically, an immense number of quantifiable imaging features, are extracted with high-throughput computing from medical images, and correlated with other clinical data, for unraveling the complex underlying pathophysiological mechanisms which are potentially reflected in tissue macroscopic phenotype. Therefore, radiomic analysis is holding a key role in the future of clinical decision support, as it provides powerful tools for assessing the spatiotemporal heterogeneity of breast cancer, aiming at increasing the diagnostic accuracy and predicting patient's outcome toward individualized therapy planning.
Machine learning based on multi-parametric magnetic resonance imaging to differentiate glioblastoma multiforme from primary cerebral nervous system lymphoma
2018, European Journal of Radiology
To evaluate the performance of a machine learning method based on texture features in multi-parametric magnetic resonance imaging (MRI) to differentiate a glioblastoma multiforme (GBM) from a primary cerebral nervous system lymphoma (PCNSL).
We included 70 patients who underwent contrast enhanced brain MRI at 3 T with brain tumors diagnosed as GBM (n = 45) and PCNSL (n = 25) in this retrospective study. Twelve histograms and texture parameters were assessed on T2-weighted images (T2WIs), apparent diffusion coefficient maps, relative cerebral blood volume (rCBV) map, and contrast-enhanced T1-weighted images (CE-T1WIs). A prediction model was developed using a machine learning method (univariate logistic regression and multivariate eXtreme gradient boosting-XGBoost) and the area under the receiver operating characteristic curve of this model was calculated via 10-fold cross validation. In addition, the performance of the machine learning method was compared with the judgments of two board certified radiologists.
With the univariate logistic regression model, the standard deviation of rCBV offered the highest AUC (0.86), followed by mean value of rCBV (0.83), skewness of CE-T1WI (0.78), mean value of CET1 (0.78), and max value of rCBV (0.77). The AUC of the XGBoost was significantly higher than the two radiologists (0.98 vs. 0.84; p < 0.01 and 0.98 vs. 0.79; p < 0.01, respectively).
The performance of machine learning based on histogram and texture features in multi-parametric MRI was superior to that of conventional cut-off method and the board certified radiologists to differentiate a GBM from a PCNSL.

View all citing articles on Scopus

: This work was supported in part by grant R01 CA90437 (O. Nalcioglu) and R21 CA121568 (M.-Y.Su) from the National Cancer Institute (Bethesda, MD), grants 9WB-0020 (M.-Y. Su) and 14GB-0148 (K. Nie) from the California Breast Cancer Research Program (Oakland, CA), and Cancer Center Support Grant No. 2P30CA062203-13S (F.L. Meyskens, Jr) from the National Cancer Institute (Bethesda, MD).

View full text

Original investigationPrediction of Malignant Breast Lesions from MRI Features: A Comparison of Artificial Neural Network and Logistic Regression Techniques

Rationale and Objectives

Materials and Methods

Results

Conclusion

Section snippets

Malignant and Benign Lesion Database

ANN Diagnostic Feature Selection and Evaluation

Discussion

Acad Radiol

Eur J Surg Oncol

Ultrasound Med Biol

J Clin Epidemiol

Acad Radiol

Acad Radiol

Artif Intell Med

J Clin Epidemiol

Automatic identification and classification of characteristic kinetic curves of breast lesions on DCE-MRI

Med Phys

Computerized interpretation of breast MRI: investigation of enhancement-variance dynamics

Med Phys

Breast lesion analysis of shape technique: semiautomated vs. manual morphological description

J Magn Reson Imaging

Breast MRI lesion classification: improved performance of human readers with a backpropagation neural network computer-aided diagnosis (CAD) system

J Magn Reson Imaging

Utility of magnetic resonance imaging in the management of breast cancer: evidence for improved preoperative staging

J Clin Oncol

Breast carcinoma: effect of preoperative contrast-enhanced MR imaging on the therapeutic approach

Radiology

Staging of symptomatic primary breast cancer with MR imaging

AJR Am J Roentgenol

Pre-operative staging of invasive breast cancer with MR mammography and/or PET: boon or bunk?

Br J Radiol

The role of contrast-enhanced MR mammography for determining candidates for breast conservation surgery

Breast Cancer

Computerized analysis of breast lesions in three dimensions using dynamic magnetic-resonance imaging

Med Phys

Computerized classification of malignant and benign microcalcifications on mammograms: texture analysis using an artificial neural network

Phys Med Biol

Comparison of artificial neural networks with other statistical approaches

Cancer

Original investigation
Prediction of Malignant Breast Lesions from MRI Features: A Comparison of Artificial Neural Network and Logistic Regression Techniques