Unsupervised Deep Learning Applied to Breast Density Segmentation and Mammographic Risk Scoring

0
834
Unsupervised deep learning applied to breast density segmentation and mammographic risk scoring

Unsupervised Deep Learning Applied to Breast Density Segmentation and Mammographic Risk Scoring

Abstract

Unsupervised deep learning applied to breast density segmentation

Unsupervised deep learning applied to breast density segmentation and mammographic risk scoring.We present a method that learns a feature hierarchy from unlabeled data.The proposed model learns features at multiple scales. We evaluated our method on three different clinical datasets.
Our state-of-the-art results show that the learned breast density scores have a very strong positive relationship with manual ones, and that the learned texture scores are predictive of breast cancer.
The model is easy to apply and generalizes to many other segmentation and scoring problems.Tamoxifen, for example, reduces breast density and decreases the risk, whereas hormone replacement therapy causes the opposite
Many MD scores have been proposed, ranging from manual categorical (e.g., BI-RADS) to automated continuous scores.In Cumulus, the radiologist sets an intensity threshold to separate radiodense (white appearing) from fatty (dark appearing) tissue. The computer then measures the proportion of dense to total breast area, known as percentage mammographic density (PMD).
However, user-assisted thresholding is subjective and time-consuming, and hence not suited for large epidemiological studies.

Conclusion

We have presented an unsupervised feature learning method for breast density segmentation and automatic texture scoring.

 The model learns features across multiple scales.

Once the features are learned, they are fed to a simple classifier that is specific to the task of interest.

The results suggest that the proposed method was able to learn useful features for each of the considered applications.

The automated MT scores separated cancers and controls better than two state-of-the-art MT scoring methods.

In the texture dataset the CSAE model improved on the KNN method by Nielsen et al. and a simplified version of the model of Häberle et al. The full model of Häberle et al. could not be tested, as necessary parameter settings were missing.