Automated clinical image segmentation

Coronary artery disease (CAD) is one of the top four chronic disease groups and number one in terms of mortality worldwide, with approximately 50,000 deaths (Source: per year. Among them CAD is caused by blockage inside the blood vessels as a factor in major cardiovascular diseases. The process of determining the risk of CAD is very important. The Medysapiens’s solution confirms the risk from cardiovascular coronary X-ray angiography. The process consists of four steps, as follows

1) Cardiovascular segmentation step:

Lad (Left Anterior Descending artery), LCX (Left Circumflex Artery), RCA (Right Coronary Artery) segmentation work to make up the coronary artery from the coronary artery X-ray angiography taken in different directions (The image above is LCX and the X-ray image is not the original).

2) Lesion detection step:

Secondly, using a deep learning model performs the operation of detecting the location of the narrowed area of the main blood vessel, that is, the site of the suspected lesion.

SYNTAX score is divided into high (≤33), medium (23-32), low (<22) as a specified indicator to quantitatively assess the risk of cardiovascular disease in the American Heart Association (AHA), the high and medium level is corresponding to the CABG(Coronary Artery Bypass Graft) and low level is corresponding to the PCI (Percutaneous Coronary Intervention).

Improves the accuracy of the system by eliminating false positives by adding a process that uses image post-processing to determine whether the detected lesion is a true lesion. Specifically, the location of the lesion is identified through the detection of the lesion, the skeleton of the blood vessel is formed at the location, and the distance to the inner wall of the skeleton is measured to distinguish the actual lesion.

3) Left Main (LM) segmentation step:

Thirdly, we perform a segmentation operation for extracting the left main entrance to the left main vessels (LM) that has a significant impact on the SYNTAX score.

LM refers to the area where the blood vessels of LAD and LCX start, and the area that shows the greatest risk in the SYNTAX score if there is a lesion in the blood vessels in the LM region. For this reason, LAD and LCX use the deep learning model to segment the LM area to make a more accurate judgment.

4) SYNTAX score classification step:

The input used in the classification of the SYNTAX score is as follows: The SYNTAX score for the vessels of the patient is classified by the characteristics of the vessel image, the LM image, and the heatmap images that detected lesions by the deep learning model.

Without simply adopting a method of classifying the SYNTAX score from X-ray angiography, segmentation of the LM region and the entire blood vessel and detect the location of the lesion and classify the SYNTAX score based on this. This provides cardiologists with coronary artery evaluation auxiliary solutions with the interpretability of a high detail process as well as improvements in performance.