Monolithic scintillation detectors for positron emission tomography and single-photon emission computed tomography (SPECT) imaging have many advantages over pixelated detectors. The use of monolithic crystals allows for reducing the scintillator cost per unit volume and increasing the sensitivity along with the energy and timing resolution of the detector. In addition, on thick detectors the depth-of-interaction can be determined without additional hardware. However, costly and complex calibration procedures have been proposed to achieve optimal detector performance for monolithic detectors. This hampers their use in commercial systems. There is thus, a need for simple calibration routines that can be performed on assembled systems. The main goal of this work is to develop a simplified calibration procedure based on acquired training data. In comparison with other methods that use training data acquired with beam sources attached to robotic stages, the proposed method uses a static un-collimated activity source with simple geometry acquiring in a reasonable time. Once the data are acquired, the calibration of the detector is accomplished in three steps: energy calibration based on the k-means clustering method, self-organization based on the self-organizing maps algorithm, and distortion correction based on the Monge–Kantorovich grid adaptation. The proposed calibration method was validated for 2D positioning using a SPECT detector. Similar results were obtained by comparison with an existing calibration method (maximum likelihood estimation). In conclusion, we proposed a novel calibration method for monolithic scintillation detectors that greatly simplifies their use with optimal performance in SPECT systems.