Selection of diagnostic features on breast MRI to differentiate between malignant and benign lesions using computer-aided diagnosis: differences in lesions presenting as mass and non-mass-like enhancement
Related Experiment Video
Updated: Jun 19, 2026
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Dustin Newell1, Ke Nie, Jeon-Hor Chen
1Tu & Yuen Centre for Functional Onco-Imaging (CFOI), University of California, Irvine, CA 92697-5020, USA.
Purpose:
To investigate methods developed for the characterisation of the morphology and enhancement kinetic features of both mass and non-mass lesions, and to determine their diagnostic performance to differentiate between malignant and benign lesions that present as mass versus non-mass types.
Methods:
Quantitative analysis of morphological features and enhancement kinetic parameters of breast lesions were used to differentiate among four groups of lesions: 88 malignant (43 mass, 45 non-mass) and 28 benign (19 mass, 9 non-mass). The enhancement kinetics was measured and analysed to obtain transfer constant (K(trans)) and rate constant (k(ep)). For each mass eight shape/margin parameters and 10 enhancement texture features were obtained. For the lesions presenting as nonmass-like enhancement, only the texture parameters were obtained. An artificial neural network (ANN) was used to build the diagnostic model.
Results:
For lesions presenting as mass, the four selected morphological features could reach an area under the ROC curve (AUC) of 0.87 in differentiating between malignant and benign lesions. The kinetic parameter (k(ep)) analysed from the hot spot of the tumour reached a comparable AUC of 0.88. The combined morphological and kinetic features improved the AUC to 0.93, with a sensitivity of 0.97 and a specificity of 0.80. For lesions presenting as non-mass-like enhancement, four texture features were selected by the ANN and achieved an AUC of 0.76. The kinetic parameter k(ep) from the hot spot only achieved an AUC of 0.59, with a low added diagnostic value.
Conclusion:
The results suggest that the quantitative diagnostic features can be used for developing automated breast CAD (computer-aided diagnosis) for mass lesions to achieve a high diagnostic performance, but more advanced algorithms are needed for diagnosis of lesions presenting as non-mass-like enhancement.
