Using artificial intelligence to model expert panel diagnosis of cholecystitis severity

Area of Science:

• Medical imaging
• Artificial intelligence in surgery
• Surgical outcomes prediction

Background:

• The Parkland Grading Scale (PGS) assesses cholecystitis severity, aiding in predicting surgical difficulty and complications.
• Expert panel consensus reduces subjectivity in PGS grading but is time-intensive.
• AI models offer potential for efficient and consistent image-based diagnostic assessments.

Purpose of the Study:

• To develop and evaluate AI models for automated cholecystitis severity grading using the PGS.
• To compare AI model performance against expert surgical panel consensus.
• To assess the interpretability of AI models in grading cholecystitis severity.

Main Methods:

• Laparoscopic cholecystectomy videos were analyzed, with representative frames manually graded by three surgical experts.
• Inter-rater variability was assessed using weighted Cohen's kappa.
• Two AI models were developed for automated PGS grading and their accuracy and interpretability evaluated.

Main Results:

• Expert panel consensus achieved high inter-rater reliability (kappa: 0.76-0.83).
• AI Model B demonstrated higher accuracy (72%, kappa: 0.77) compared to Model A (69%, kappa: 0.62) against the expert consensus.
• Model B's grading was influenced by gallbladder, liver, and omentum appearance.

Conclusions:

• A transformer-based AI model can effectively predict PGS ratings, performing comparably to individual experts.
• The inherent subjectivity and variance in the PGS present limitations for its use as a definitive ground truth in AI development.
• AI shows promise in standardizing cholecystitis severity assessment, but further refinement is needed.