Automatic neoplasms detection in mammograms

By | 05.03.2018

High people morbidity is a problem for most regions. Moreover, some diseases that need early diagnostics can be discovered too late. Thereby, the government considers introducing of mass screening system (installing plenty of X-ray machines) to provide regular examination for as much percentage of population as possible. But there is another problem: a lack of qualified medical personnel for the amount of equipment that already present.

Detection system of anomalies in mammograms is “PAWLIN Technologies” development which was presented on “Best medical IT-system 2013” contest. Of course, automatic systems cannot replace a medical specialist but can found that a patient has a problem and really have to consult a doctor. Thereby, such diagnostics can reduce treatment costs and increase the percentage of cured patients.

Our company developed a training sample and, as a result, an AI that’s able to examine mammogram. In case if potential problem is found AI warns about consulting a doctor. Hybrid neural network model detects primary marks using automatic vision and transmits these marks to a neural network that classificate mammograms.

Left picture — X-ray image with neoplasm

Right picture — X-ray image without neoplasm

To test the detection quality we used 20 images with neoplasms and 69 images without neoplasms. Positive and negative examples are taken from training images library. These examples are divided into training and testing samples in a ratio of 50% to 50%.

Process of objects detection:

  1. Application of neural network classifier file obtained after training process
  2. Extraction of potential areas and marks for classification
  3. Classification objects using neural network classifier
  4. Visualization of detected objects

Also we must pay attention to human factor: the neoplasms that are not detected for any reasons by a doctor can cause patient’s death as it is highly dangerous not to find neoplasms at an early stage.

Left picture — our algorithm

Right picture — examination made by an expert

Reached accuracy rate is 90% that shows high quality of developed detection algorithm. The results are at the level of neoplasms detection made by an expert.

We have received cooperation proposals from different commercial organizations and leading universities in this field.

“PAWLIN Technologies” is looking for partners among medical organizations and medical equipment manufacturers which are interested in applying this technology and have a large archive of classified X-ray images for cooperation in developing of classifiers for related diseases.