A method and apparatuses are provided to identify and correct partial volume errors (PVEs) in material decomposition of a spectral computed tomography (CT) scan, due to different X-ray trajectories incident on a same macro-pixel passing through different material components (e.g., bone and water). Macro-pixels are virtual crystals generated by aggregating the signals/counts from several smaller actual pixels (i.e., micro-pixels) of a detector array. Thus, when a PVE is identified within a macro-pixel, the separate signals/counts from the micro-pixels can be used for material decomposition, instead of the aggregated signals/counts of the macro-pixel, thereby providing improved spatial resolution of the material components and, at least partial, overcoming the PVE. A measure of the difference between spectrally-resolved counts based a material projection lengths (e.g., from a calibrated lookup table) and the measured counts of the macro-pixel can be used to identify PVEs, e.g., when the difference measure exceeds a predefined threshold.