AI Driven National Platform for CT cOronary Angiography for clinicaL and industriaL applicatiOns Registry

AI precision toolkits: AI stenosis reporting

Stenosis reporting: Severity of stenosis and accurate anatomical localization of stenosis. The significance of a stenosis is determined by visual estimation of the maximal grade of luminal narrowing caused by the plaque. As recommended in SCCT guideline (Leipsic et al., 2014) , coronary stenosis can be graded as minimal, mild, moderate, severe and total occluded separately. Following the guideline, a stenosis will be classified as obstructive and non-obstructive. The location of the stenosis uses the SCCT model (Leipsic et al., 2014)

AI precision toolkits: Agatston scoring

Agatston scoring: Agatston scoring of calcified plaque. As recommended in SCCT clinical practical guidelines (Leipsic et al., 2014), Agatston scoring programs generally identify pixels that exceed 130 HU as a level corresponding to calcium on a non-contrast study (Agatston et al., 1990) . The reader needs to identify each lesion discrete calcific focus) in each vessel distribution. The summed score for each vessel is generated by the scoring program based on an area-density (Agatston score) (Agatston et al., 1990) measurement of each calcified focus. The total coronary Agatston score is the sum of all calcified lesions in all coronary beds.

AI precision toolkits: Plaque

Plaque analysis: Plaque volume, burden, type and anatomical locations. Coronary segmentation and plaque analysis is performed for segments with diameter ≥1.5 mm. Location of plaque uses the SCCT model (Leipsic et al., 2014). For each plaque, the reader marks its start-and end-points, quantifies plaque area,volume and plaque burden, and specifies its type (non-calcified, calcified, or mixed) (Achenbach et al., 2004) . Additionally, non-calcified plaque can be further divided into low attenuation plaque (LAP). A HU \<30 will signify LAP and \>30 will signify non-LAP.

AI precision toolkits: EAT analysis

EAT analysis: Total volume and anatomical locations. EAT and pericardial adipose tissue (PAT) are metabolically active fat surrounding the coronary artery and the heart, being associated with increased risk of cardiovascular disease (Villasante et al, 2019) . EAT can be quantified on non-contrast CT scans. The annotations on the CT scans are obtained by manually drawing the pericardium first to define the region. EAT is identified using the adipose tissue attenuation references between -190 and -30 HU (Oikonomou et al., 2018) . Due to the CT scan noise and changing of attenuation, the HU value of fat can vary, so the final EAT region is verified by an experienced radiologist or cardiologist.

AI outcome analysis

1\. Mortality (All-cause and/or cardiovascular)

AI outcome analysis

Major-adverse cardiovascular events (myocardial infarction, stroke, heart failure, revascularisation, arrhythmias, etc)

AI outcome analysis

Re-hospitalisation