AutoCumulus: an automated mammographic density measure created using artificial intelligence

Mammographic (or breast) density is an established risk factor for breast cancer, previously measured using a variety of quantitative, semi-automated and automated approaches. We present a new automated measure, AutoCumulus, learned from applying deep learning to semi-automated measures.

We studied the mammograms of 9,057 population-screened women in the BRAIx program for which semi-automated measurements of mammographic density had been made by experienced readers using the CUMULUS software. The dataset was split into training, testing, and validation sets (80%, 10%, and 10%, respectively). We applied a deep learning regression model (fine-tuned ConvNeXtSmall) to estimate percentage density and assessed performance by the correlation between estimated and measured percent density using the testing dataset. The automated measure was independently tested using the CSAW-CC dataset in which density was measured using the LIBRA software by comparing measures for the left and right breasts, and the specificity for high sensitivity and the area under the receiver operating characteristic curve (AUC) for interval cancers.

The correlation in percent density between the automated and human measures was 0.95. Based on the CSAW-CC dataset, AutoCumulus outperformed LIBRA in terms of the correlation between the left and right breast (0.95 versus 0.79; P < 0.001), specificity for 95% sensitivity (13% versus 10%; P < 0.001 using McNemar's test), and AUC (0.638 versus 0.597; P < 0.01 using DeLong's test) for interval cancers.

We have created an automated measure of mammographic density that is highly accurate and allows rapid measurement of large numbers of mammograms. Compared with a well-established automated measure, AutoCumulus showed higher within-woman repeatability and modestly better prediction of interval cancers. These findings support its potential value as a scalable risk indicator, although further validation is required before clinical implementation.