Purpose

This study investigated the impact of arms/hands and body position on the measurement of breast density using MRI.

Methods

Noncontrast-enhanced T1-weighted images were acquired from 32 healthy women. Each subject received four MR scans using different experimental settings, including a high resolution hands-up, a low resolution hands-up, a high resolution hands-down, and finally, another high resolution hands-up after repositioning. The breast segmentation was performed using a fully automatic chest template-based method. The breast volume (BV), fibroglandular tissue volume (FV), and percent density (PD) measured from the four MR scan settings were analyzed.

Results

A high correlation of BV, FV, and PD between any pair of the four MR scans was noted (r > 0.98 for all). Using the generalized estimating equation method, a statistically significant difference in mean BV among four settings was noted (left breast, score test p = 0.0056; right breast, score test p = 0.0016), adjusted for age and body mass index. Despite differences in BV, there were no statistically significant differences in the mean PDs among the four settings (p > 0.10 for left and right breasts). Using Bland-Altman plots, the smallest mean difference/bias and standard deviations for BV, FV, and PD were noted when comparing hands-up high vs low resolution when the breast positions were exactly the same.

Conclusions

The authors' study showed that BV, FV, and PD measurements from MRI of different positions were highly correlated. BV may vary with positions but the measured PD did not differ significantly between positions. The study suggested that the percent density analyzed from MRI studies acquired using different arms/hands and body positions from multiple centers can be combined for analysis.

Objectives

To apply deep learning algorithms using a conventional convolutional neural network (CNN) and a recurrent CNN to differentiate three breast cancer molecular subtypes on MRI.

Methods

A total of 244 patients were analyzed, 99 in training dataset scanned at 1.5 T and 83 in testing-1 and 62 in testing-2 scanned at 3 T. Patients were classified into 3 subtypes based on hormonal receptor (HR) and HER2 receptor: (HR+/HER2-), HER2+, and triple negative (TN). Only images acquired in the DCE sequence were used in the analysis. The smallest bounding box covering tumor ROI was used as the input for deep learning to develop the model in the training dataset, by using a conventional CNN and the convolutional long short-term memory (CLSTM). Then, transfer learning was applied to re-tune the model using testing-1(2) and evaluated in testing-2(1).

Results

In the training dataset, the mean accuracy evaluated using tenfold cross-validation was higher by using CLSTM (0.91) than by using CNN (0.79). When the developed model was applied to the independent testing datasets, the accuracy was 0.4-0.5. With transfer learning by re-tuning parameters in testing-1, the mean accuracy reached 0.91 by CNN and 0.83 by CLSTM, and improved accuracy in testing-2 from 0.47 to 0.78 by CNN and from 0.39 to 0.74 by CLSTM. Overall, transfer learning could improve the classification accuracy by greater than 30%.

Conclusions

The recurrent network using CLSTM could track changes in signal intensity during DCE acquisition, and achieved a higher accuracy compared with conventional CNN during training. For datasets acquired using different settings, transfer learning can be applied to re-tune the model and improve accuracy.

Key points

• Deep learning can be applied to differentiate breast cancer molecular subtypes. • The recurrent neural network using CLSTM could track the change of signal intensity in DCE images, and achieved a higher accuracy compared with conventional CNN during training. • For datasets acquired using different scanners with different imaging protocols, transfer learning provided an efficient method to re-tune the classification model and improve accuracy.