PT - JOURNAL ARTICLE AU - Lafontaine, Daniel AU - Schmidtlein, Charles AU - Kesner, Adam AU - Kirov, Assen AU - O'Donoghue, Joseph AU - Humm, John TI - <strong>Integration of open source tools with applications to machine learning segmentation in the open source TriDFusion (3DF) medical image viewer software</strong> DP - 2023 Jun 01 TA - Journal of Nuclear Medicine PG - P495--P495 VI - 64 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/64/supplement_1/P495.short 4100 - http://jnm.snmjournals.org/content/64/supplement_1/P495.full SO - J Nucl Med2023 Jun 01; 64 AB - P495 Introduction: The TriDFusion (3DF) (TriDFusion (3DF) image viewer. EJNMMI Phys 9, 72 (2022). https://doi.org/10.1186/s40658-022-00501-y) software is an open source medical imaging viewer developed at our institution, which has recently been benchmarked and released. The software was developed to address a broad array of unmet needs for the nuclear medicine research community by providing a flexible image viewer with analysis tools that are not commonly available in other opensource and commercial nuclear medicine software platforms. 3DF is opensource software that provides many integrated tools for reviewing and analyzing multi-modality images. In this study we demonstrate the ability of 3DF to integrate external open source tools developed for CT segmentation. Methods: A powerful open source python based deep learning organ CT segmentation tool developed independently at Cornel University called "TotalSegmentator" by Wassthal et al (Wasserthal J., Meyer M., Breit H., Cyriac J., Yang S., Segeroth M. TotalSegmentator: robust segmentation of 104 anatomical structures in CT images, 2022. URL: https://arxiv.org/abs/2208.05868. arXiv: 2208.05868), was integrated into 3DF. Local graphical processing units were leveraged to efficiently perform the segmentation in a user-friendly manner. Several sample patient images were processed and analyzed. Results: The 3DF software’s existing image viewing framework facilitated the integration of these tools. Once integrated, the machine learning segmentation tool allows a researcher to automatically segment multiple organs into data structures that can be used in a variety of tasks. The results can be exported as .csv object statistics file, DICOM RT structure or DICOM mask. Examples will be shown of lung-liver ratio on a SPECT/CT imag and automatic 3D lung lobe segmentation. These automated processing applications are carried out with a single user mouse click.Conclusions: The 3DF presents an accessible image analysis software option for the medical imaging research community. In this work we demonstrated the ability of 3DF to efficiently integrate new , externally developed capabilities, specifically machine learning-based CT segmentation. This open source initiative can help users integrate evolving technology and/or develop and share their own custom image processing applications. Furthermore, 3DF’s user friendly interface provides new tools to clinical researchers who lack the time required to create custom research workflows. Promising applications of this 3DF software are organ segmentation for radionuclide dosimetry, 3D nuclear medicine analysis such as lung shunt fraction and lobe quantification etc.