Skychain AI for Breast Cancer Recognition released

Skychain ANN for recognition of breast cancer on mammograms is now available in Skychain Alpha at no cost within the framework of Breast Cancer Awareness month!

Breast cancer is an uncontrolled growth of breast cells. To better understand breast cancer, it helps to understand how any cancer can develop.

Cancer occurs as a result of mutations, or abnormal changes, in the genes responsible for regulating the growth of cells and keeping them healthy. The genes are in each cell’s nucleus, which acts as the “control room” of each cell. Normally, the cells in our bodies replace themselves through an orderly process of cell growth: healthy new cells take over as old ones die out. But over time, mutations can “turn on” certain genes and “turn off” others in a cell. That changed cell gains the ability to keep dividing without control or order, producing more cells just like it and forming a tumor.

Breast cancer — Symptoms and treatment

Breast cancer is always caused by a genetic abnormality (a “mistake” in the genetic material). However, only 5–10% of cancers are due to an abnormality inherited from your mother or father. Instead, 85–90% of breast cancers are due to genetic abnormalities that happen as a result of the aging process and the “wear and tear” of life in general.

Breast cancer is the first among the known cancer cases, according to Cancer Statistics Center of American Cancer Society.

The problem

Age-standardised death rates from Breast cancer by country (per 100,000 inhabitants) © Wikipedia.org

One in eight women are expected to develop breast cancer during their lifetimes. Routine mammograms can help catch breast cancer early, but reading mammograms is a labour-intensive and time-consuming task. Because of relatively high error rates and the large variance between radiologist readings, the current breast cancer diagnosis protocol requires double-blind reads by two independent radiologists.

The American Cancer Society’s estimates for breast cancer in the United States for 2018 are:

About 266,120 new cases of invasive breast cancer will be diagnosed in women.

About 63,960 new cases of carcinoma in situ (CIS) will be diagnosed (CIS is non-invasive and is the earliest form of breast cancer).

About 40,920 women will die from breast cancer this year.

Breast cancer is the second leading cause of cancer death in women (only lung cancer kills more women each year). The chance that a woman will die from breast cancer is about 1 in 38 (about 2.6%).

The Skychain solution

Breast tumors differ in the degree of malignancy and histological signs. Because of the variety of tumor manifestations, its segmentation is a challenging task.

The Skychain neural network is capable of recognizing tumor tissues on mammograms and determining the type of tumor (1 of 4):

  • calc benign
  • mass benign
  • calc malignant
  • mass malignant.

For the ANN training, we used data set of the Curated Breast Imaging Subset for the Digital Database for Screening Mammography, which contains 2620 scans of high-resolution X-rays (10–15 mega-pixels each, 160 GB images in total).

To speed up the training process on such a big number of images, we used downsampling by sampling 256 patches of 224x224 pixels from each image.

There were different data patch allocations for each mode:

  • for segmentation of tumor tissues: 70% and 30% — undamaged/damaged tissues.
  • For classification of damaged areas: 30% — undamaged tissues; 1518% for each class of damaged tissues.

The Convolutional Neural Network, built on the basis of such popular architectures as ShuffleNet, MobileNetV2, NASNetMobile and DenseNet121 was used.

On the left is an image with true segmentation, on the right is a segmentation predicted by the developed neural network.

Below are ROC curves for the recognition of different tumor types in detailed segmentation.

Watch the interview on the Skychain AI for breast cancer recognition with its developer Pavel Maevskikh:

The future

As breast cancer is more easily diagnosed with each year, we believe that the percent of survival may even increase with our ANN available for use. But the fact that the diagnostics is getting more and more accurate doesn’t mean people should forget about the risk prevention.

To lower the risk of breast cancer one can follow these simple steps:

Get to and stay at a healthy weight: Both increased body weight and weight gain as an adult are linked with a higher risk of breast cancer after menopause.

Be physically active: Many studies have shown that moderate to vigorous physical activity is linked with lower breast cancer risk, so it’s important to get regular physical activity.

Limit or avoid alcohol: Alcohol also increases risk of breast cancer. Even low levels of alcohol intake have been linked with an increase in risk.

Other factors that might lower risk: Women who choose to breastfeed for at least several months may also get an added benefit of reducing their breast cancer risk.

Join Skychain on social media: Twitter, Facebook, Telegram

Try Skychain Alpha.

Egor Chertov, Skychain team