A Modular DICOM Framework for Imaging Informatics: Metadata Management and AI-Assisted Anatomical Consistency Review

Medical imaging workflows require reliable DICOM processing, metadata management, communication, data protection, and quality control. This study presents a modular Python desktop framework integrating image visualization, DICOM communication, controlled working copies, and human-supervised anatomical consistency review. The documented model-development dataset comprised 4846 MRI DICOM instances distributed across 21 anatomical and technical categories from one operational environment and one manufacturer. The DenseNet121 checkpoint contained 30 output nodes; the deployed application mapped 13 anatomical categories and grouped the remaining indices as Unknown. Mapped image-derived predictions were compared primarily with normalized BodyPartExamined values. To examine the response to deliberately discordant metadata, a paired functional experiment compared original and modified copies of five MRI studies, comprising 31 imaging entries per condition. Functional testing confirmed recursive import, hierarchical organization, pseudonymization of working copies, C-ECHO, and C-STORE reception and transmission. The software handled studies containing more than 2000 instances, with initial loading times below 3 s and memory consumption below 1.2 GB in the evaluated environment; DICOM reception reached approximately 85 Mbps on a local Gigabit Ethernet network. All 31 final anatomical predictions were unchanged between paired conditions. The interface displayed 10 alerts with original metadata and 31 with modified metadata, corresponding to 21 newly displayed alerts; 10 alerts were already present before modification. The findings document the framework's technical feasibility and its functional response to controlled metadata discordance. The five-study experiment does not establish external classification performance or clinically validated metadata-error detection.