Automated selection of hypointense regions in diffusion-weighted breast MRI

Recent research suggests that diffusion-weighted (DW) MRI, and in particular the apparent diffusion coefficient (ADC), can be used to improve the sensitivity and specificity of dynamic contrast-enhanced (DCE) MRI for the detection of breast cancer. However, to date the methods proposed for determining a representative ADC value for a suspicious lesion are highly varied. One approach is to compute the mean ADC value over the entire lesion to obtain a representative ADC value. Another is to compute the mean ADC value within one or more regions of interest (ROIs) defined on the suspicious lesion. The earliest examples of this approach involve manually defining ROIs of hypointensity to be as large as possible, but constrained within the lesion, and such that areas of necrosis are avoided in large lesions. More recent examples of this approach involve placing one or more smaller ROIs of hypointensity within a suspicious lesion and computing, for example, the global minimum [1] or mean [2]. This latter approach appears to provide better discrimination between benign and malignant lesions. Nevertheless to date there does not exist a well-defined and objective method for defining these ROIs. The problem is complicated by the typically low signal-to-noise ratio in the DW images. We propose an automated method based on the converging squares algorithm [3], which is a multiscale minimum finding technique with inherent robustness to noise. We also present an evaluation of the method, using routine clinical data, for computing a representative ADC value for discriminating benign and malignant lesions. The method is also compared to ensemble averaging of ADC values over the entire lesion and the selection of the global minimum ADC value.