Prediction of Post-Infectious Inflammatory Response Syndrome in Patients with Cryptococcal Meningitis Based on Clinical and Radiomics Data

Background/Objectives: Post-infectious inflammatory response syndrome (PIIRS) is a rare complication of cryptococcal meningitis. The objective of this study was to identify predictors that can be used to identify patients at risk of PIIRS before its onset. Methods: A total of 149 patients with cryptococcal meningitis who did not develop PIIRS (controls, n = 84) and developed PIIRS (cases, n = 65) were included in the study. Clinical presentation data, treatment data, and laboratory data at admission were collected and compared between the two groups using univariate and multivariable analyses. Inflamed areas within T2 FLAIR MRI scans were manually annotated, and a total of 110 radiomic features were extracted per patient. Data were split into training, test, and validation sets, and multiple machine learning models integrating radiomic, spatial, and clinical features were developed and evaluated using area under the curve (AUC) and related performance metrics. Results: Ventriculoperitoneal shunt (odds ratio: 4.64 [1.84–12.46]; p-value = 0.001), complement component 3 (C3) (odds ratio: 5.78 [1.55–24.42]; p-value = 0.012), and lumbar puncture opening pressure (odds ratio: 1.01 [1.00–1.01]; p-value = 0.008) were independent risk factors associated with PIIRS development. All four radiomics models performed very well in predicting PIIRS, with the final model (region of interest + location + C3) achieving the best overall performance set (test set AUC: 0.948 [0.840–1.000], validation set AUC: 0.943 [0.814–1.000]). Conclusions: Radiomics-based classification models demonstrated good performance for predicting PIIRS. These findings should be considered hypothesis-generating given the small sample size and require validation in larger multicenter studies before clinical implementation.