Guided review by frequent itemset mining: additional evidence for plaque detection

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Data Mining

Background:

• Manual detection of coronary plaque in CTCA can be challenging, leading to potential missed diagnoses.
• Observer variability is a significant concern in interpreting CTCA scans for coronary artery disease.

Purpose of the Study:

• To develop and evaluate a guided review process for manual coronary plaque detection in CTCA.
• To leverage frequent itemset mining for learning spatial plaque distribution patterns.
• To predict and identify potentially missed plaques during CTCA analysis.

Main Methods:

• Utilized plaque distribution patterns from 252 CTCA patient datasets.
• Employed frequent itemset mining to learn spatial plaque distribution rules from training data.
• Applied learned rules to identify segments with potential unlabeled plaques for guided review.
• Compared the guided review approach against a weighted random approach.

Main Results:

• The guided review process significantly outperformed the weighted random approach (p < 0.001) in predicting segments with missed coronary plaques.
• Up to 47% of initially missed plaques were rediscovered by reviewing only 4.4% of all segments.
• Demonstrated the efficacy of spatial plaque patterns in guiding detection.

Conclusions:

• Spatial distribution patterns of coronary atherosclerosis, identified via frequent itemset mining, can effectively predict missed plaques.
• The proposed guided review method shows potential to reduce both intra- and inter-observer variability in CTCA interpretation.
• This approach enhances the accuracy and efficiency of coronary plaque detection.