AI-Assisted Simple Scoring Algorithm Was Helpful in the Risk Assessment of Cardiac Involvement in Patients with Pulmonary Sarcoidosis

<b>Background:</b> Cardiac involvement, one of the most life-threatening complications of sarcoidosis, remains under-recognized due to its oligo-symptomatic presentation in some patients. This retrospective study aimed to evaluate the utility of various clinical predictors of cardiac sarcoidosis (CS) development. <b>Methods:</b> The study included patients with pulmonary sarcoidosis diagnosed according to the recent ATS guidelines between January 2020 and July 2024 who underwent cardiac magnetic resonance (CMR) due to clinical suspicion of CS. The original Lake Louise criteria were used to identify active myocarditis. <b>Results:</b> Out of 393 patients diagnosed with pulmonary sarcoidosis, CMR was performed in 92 patients. Cardiac sarcoidosis was confirmed in 48 patients (52%, CS+), and excluded in 44 patients (48%, CS-). CS(+) patients demonstrated significantly more frequent Holter ECG abnormalities and liver/spleen sarcoidosis compared to CS(-) patients. Stage IV pulmonary disease, ECG abnormalities, and hypercalcemia were more common in CS(+) than in CS(-) patients; however, these differences did not reach statistical significance. Multivariate analysis identified Holter ECG abnormalities and liver/spleen involvement as significant predictive factors for CS, increasing the risk of cardiac involvement by approximately 4- and 6-fold, respectively. An AI-assisted simple scoring system based on five parameters: ECG abnormalities, Holter ECG abnormalities, liver/spleen involvement, gender, and stage of sarcoidosis predicted CS with a sensitivity of 76% and specificity of 74%, using an optimal cut-off value of ≥7.6 points. <b>Conclusions:</b> In patients with pulmonary sarcoidosis, an AI-assisted scoring algorithm derived from L1-regularized logistic regression results accurately predicted cardiac involvement on CMR with high specificity and sensitivity. Prospective validation of this algorithm is necessary to confirm its clinical utility in predicting cardiac sarcoidosis.