Relating planar lung scintigraphic image features to bronchopulmonary anatomy is a mental task requiring specialized medical experience. This study aimed to accurately normalize spatial data to overlay patient images onto a bronchopulmonary segment atlas (BSA), enhancing image interpretation for nonexperts and enabling quantification. Methods: This study evaluates the efficacy of 3 spatial normalization techniques: naïve registration, cost function masking, and perfusion defect removal with convolutional autoencoders. Autoencoders were trained for each of 6 projection angles using a large cohort of healthy patients (n = 660). Perfusion planar population templates for each projection, with its corresponding BSA, were constructed using a random subset sample of these patients (n = 149). Synthetic perfusion defects were applied on 60 projections from 10 patients with normal perfusion, allowing a comprehensive assessment of each spatial normalization technique's performance and effect on defect size in the template space. Results: The results reveal that autoencoder preprocessing significantly outperforms the naïve method and exhibits comparable or superior performance to cost function masking, particularly in preserving defect size and minimizing registration error to the population template within the defect. Visual comparisons further support the efficacy of autoencoder preprocessing in preserving anatomic features. Conclusion: Autoencoder preprocessing is a fully automatic and reliable method for reducing distortions during spatial normalization in perfusion scintigraphy, highlighting its potential for enhancing registration accuracy in clinical practice for BSA overlay and defect quantification.