Deep learning model estimating breast density could help with predicting cancer risk
The researchers have developed a procedure that would estimate a density score upon feeding in a mammogram image as an input.
NEW DELHI: Researchers have developed a new deep learning model that can estimate breast density, which could be useful for cancer risk prediction.
The researchers from the University of Manchester, UK, said that the automatic feature extraction from the training data enabled by the deep learning-based approach makes it appealing for breast density estimations.
Breast density is defined as the proportion of fibro-glandular tissue within the breast and is often used in assessing the risk of developing breast cancer.
They have published their findings in the Journal of Medical Imaging.
The researchers used two independent deep learning models, initially trained on ImageNet, a non-medical imaging dataset with over a million images, and trained them with their medical imaging data through an approach known as "transfer learning".
Training and building deep learning models from the ground up is challenging owing to limited datasets, they said.
Experts that included radiologists, advanced practitioner radiographers, and breast physicians assigned density values on a visual analogue scale in 1,60,000 full-field digital mammogram images form 39,357 women.
Using this data, the researchers developed a procedure that would estimate a density score upon feeding in a mammogram image as an input.
The procedure involved preprocessing the images to make the training process computationally less intensive, extracting features from the processed images with the deep learning models, mapping the features to a set of density scores, and then combining the scores using an ensemble approach to produce a final density estimate.
The team thus developed highly accurate models for estimating breast density and its correlation with cancer risk, while conserving computation time and memory.
"The model's performance is comparable to those of human experts within the bounds of uncertainty," said lead researcher Susan M. Astley.
"Moreover, it can be trained much faster and on small datasets or subsets of the large dataset."
The researchers noted that the framework is not only limited to estimating breast cancer risk but also for training other medical imaging models based on its breast tissue density estimations.
Breast cancer is the most common cancer to affect women worldwide.
While various methods are available to estimate this measure, studies have shown that subjective assessments conducted by radiologists based on visual analogue scales are more accurate than any other method.