[Update] Model output calibration: We have implemented a new feature based on isotonic model calibration, which was fitted to different age bins. The "calibrated probability" of Alzheimer's disease is now estimated based on the raw CNN model output and the ratio of true positive to true negative cases in the normative calibration subcohorts.
For more information, see: Martensen, O.H., Strauß, T., Ramedani, R., Dyrba, M., (accepted). Comparison of post-hoc calibration methods for neural network likelihood scores, in: Handels, H., et al. (Eds.), Bildverarbeitung für die Medizin 2026. Springer Fachmedien Wiesbaden, pp. 443–449. https://doi.org/10.1007/978-3-658-51100-5_87