{"id":{"repo_id":"unr","oai_identifier":"oai:scholarwolf.unr.edu:11714/11889"},"canonical_url":"https://search.dev.ndltd.org/etd/unr/oai:scholarwolf.unr.edu:11714/11889","repository":{"repo_id":"unr","name":"University of Nevada - Reno","base_url":"https://scholarwolf.unr.edu/server/oai/request"},"display":{"title":"OrganixInsights: High throughput imaging and high content screening of organoids","abstract":"Advances in biomedical imaging technologies have significantly expanded the ability of researchers to study complex biological systems at cellular and subcellular resolution. In particular, three-dimensional (3D) organoid models have emerged as powerful experimental systems for investigating tissue development, disease progression, and therapeutic response. Organoids replicate key structural and functional characteristics of in-vivo tissues, including cellular heterogeneity, spatial organization, and extracellular matrix interactions. However, high-resolution imaging of organoids generates large volumetric datasets that require advanced computational methods for analysis. This dissertation presents a comprehensive computational framework for high-content screening and quantitative analysis of 3D organoid and biomedical imaging datasets. The proposed framework integrates deep learning models for image restoration and segmentation with a scalable web-based platform for managing and analyzing microscopy data. First, we introduce Deconv3D, a hybrid transformer–convolution architecture designed for volumetric microscopy restoration. The model integrates convolutional feature extraction with windowed self-attention modules to capture both local spatial features and long-range contextual relationships within volumetric datasets. A novel attention-weighted fusion (AWF) mechanism enables adaptive integration of encoder and decoder features during image reconstruction. Experimental results demonstrate that Deconv3D achieves high-quality reconstruction with a peak signal-to-noise ratio of 30.33 dB and structural similarity index of 0.875, while maintaining a compact architecture with only 2.23 million parameters. Second, we propose 3D-Organoid-SwinNet, a transformer-based architecture for segmentation of nuclei in organoid microscopy datasets. The model combines Swin Transformer encoders with multiscale decoding layers to capture both global contextual information and fine-grained spatial features. Evaluation on breast cancer organoid datasets demonstrates that the model achieves a Dice score of 94.91, outperforming several existing segmentation approaches. Third, we present MAT3D, a multi-aperture transformer architecture for volumetric biomedical image segmentation. MAT3D introduces parallel transformer modules that process multiple spatial representations of the input volume, enabling improved modeling of complex structures. A composite loss function incorporating voxel accuracy, object count consistency, and boundary distance improves segmentation fidelity. The proposed model achieves a Dice score of 95.12 and panoptic quality of 97.01 across microscopy datasets while demonstrating strong generalization to clinical imaging datasets. In addition to the development of these deep learning architectures, this dissertation introduces OrganixInsight, a web-based imaging bioinformatics platform for managing and analyzing high-content microscopy datasets. The platform integrates deep learning models for image restoration and segmentation with tools for experimental design, dataset management, morphometric feature extraction, and interactive 3D visualization. Together, these contributions provide an end-to-end computational framework for high-content screening and phenotypic profiling of organoid models. By combining advanced machine learning techniques with scalable software infrastructure, this work enables automated analysis of large microscopy datasets and provides new opportunities for studying cellular organization, tumor progression, and drug response in complex biological systems.","abstract_html":"Advances in biomedical imaging technologies have significantly expanded the ability of researchers to study complex biological systems at cellular and subcellular resolution. In particular, three-dimensional (3D) organoid models have emerged as powerful experimental systems for investigating tissue development, disease progression, and therapeutic response. Organoids replicate key structural and functional characteristics of in-vivo tissues, including cellular heterogeneity, spatial organization, and extracellular matrix interactions. However, high-resolution imaging of organoids generates large volumetric datasets that require advanced computational methods for analysis. This dissertation presents a comprehensive computational framework for high-content screening and quantitative analysis of 3D organoid and biomedical imaging datasets. The proposed framework integrates deep learning models for image restoration and segmentation with a scalable web-based platform for managing and analyzing microscopy data. First, we introduce Deconv3D, a hybrid transformer–convolution architecture designed for volumetric microscopy restoration. The model integrates convolutional feature extraction with windowed self-attention modules to capture both local spatial features and long-range contextual relationships within volumetric datasets. A novel attention-weighted fusion (AWF) mechanism enables adaptive integration of encoder and decoder features during image reconstruction. Experimental results demonstrate that Deconv3D achieves high-quality reconstruction with a peak signal-to-noise ratio of 30.33 dB and structural similarity index of 0.875, while maintaining a compact architecture with only 2.23 million parameters. Second, we propose 3D-Organoid-SwinNet, a transformer-based architecture for segmentation of nuclei in organoid microscopy datasets. The model combines Swin Transformer encoders with multiscale decoding layers to capture both global contextual information and fine-grained spatial features. Evaluation on breast cancer organoid datasets demonstrates that the model achieves a Dice score of 94.91, outperforming several existing segmentation approaches. Third, we present MAT3D, a multi-aperture transformer architecture for volumetric biomedical image segmentation. MAT3D introduces parallel transformer modules that process multiple spatial representations of the input volume, enabling improved modeling of complex structures. A composite loss function incorporating voxel accuracy, object count consistency, and boundary distance improves segmentation fidelity. The proposed model achieves a Dice score of 95.12 and panoptic quality of 97.01 across microscopy datasets while demonstrating strong generalization to clinical imaging datasets. In addition to the development of these deep learning architectures, this dissertation introduces OrganixInsight, a web-based imaging bioinformatics platform for managing and analyzing high-content microscopy datasets. The platform integrates deep learning models for image restoration and segmentation with tools for experimental design, dataset management, morphometric feature extraction, and interactive 3D visualization. Together, these contributions provide an end-to-end computational framework for high-content screening and phenotypic profiling of organoid models. By combining advanced machine learning techniques with scalable software infrastructure, this work enables automated analysis of large microscopy datasets and provides new opportunities for studying cellular organization, tumor progression, and drug response in complex biological systems.","abstract_has_math":false,"creators":["Sohaib, Muhammad"],"institution":null,"degree_name":null,"degree_level":"Doctorate Degree","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Parvin, Bahram"],"committee_chairs":[],"committee_members":["Alvarez Ponce, David","Zhu, Xiaoshan","Xu, Hao","Shen, Yantao","Parvin, Bahram"],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-07-27T21:46:52Z","subjects":["3D Organoids","Deep Learning","High-Content Screening","Image Segmentation"],"languages":["en_US","English"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholarwolf.unr.edu/handle/11714/11889","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Parvin, Bahram"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Alvarez Ponce, David","Zhu, Xiaoshan","Xu, Hao","Shen, Yantao","Parvin, Bahram"]},{"key":"dc:creator","label":"Author","values":["Sohaib, Muhammad"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["01/01/2026"]},{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-06-25T16:22:08Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-06-25T16:22:08Z"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Doctorate Degree"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["3D Organoids","Deep Learning","High-Content Screening","Image Segmentation"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["English"]},{"key":"dc:language.iso","label":"Language (ISO)","values":["en_US"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://scholarwolf.unr.edu/handle/11714/11889"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Advances in biomedical imaging technologies have significantly expanded the ability of researchers to study complex biological systems at cellular and subcellular resolution. In particular, three-dimensional (3D) organoid models have emerged as powerful experimental systems for investigating tissue development, disease progression, and therapeutic response. Organoids replicate key structural and functional characteristics of in-vivo tissues, including cellular heterogeneity, spatial organization, and extracellular matrix interactions. However, high-resolution imaging of organoids generates large volumetric datasets that require advanced computational methods for analysis. This dissertation presents a comprehensive computational framework for high-content screening and quantitative analysis of 3D organoid and biomedical imaging datasets. The proposed framework integrates deep learning models for image restoration and segmentation with a scalable web-based platform for managing and analyzing microscopy data. First, we introduce Deconv3D, a hybrid transformer–convolution architecture designed for volumetric microscopy restoration. The model integrates convolutional feature extraction with windowed self-attention modules to capture both local spatial features and long-range contextual relationships within volumetric datasets. A novel attention-weighted fusion (AWF) mechanism enables adaptive integration of encoder and decoder features during image reconstruction. Experimental results demonstrate that Deconv3D achieves high-quality reconstruction with a peak signal-to-noise ratio of 30.33 dB and structural similarity index of 0.875, while maintaining a compact architecture with only 2.23 million parameters. Second, we propose 3D-Organoid-SwinNet, a transformer-based architecture for segmentation of nuclei in organoid microscopy datasets. The model combines Swin Transformer encoders with multiscale decoding layers to capture both global contextual information and fine-grained spatial features. Evaluation on breast cancer organoid datasets demonstrates that the model achieves a Dice score of 94.91, outperforming several existing segmentation approaches. Third, we present MAT3D, a multi-aperture transformer architecture for volumetric biomedical image segmentation. MAT3D introduces parallel transformer modules that process multiple spatial representations of the input volume, enabling improved modeling of complex structures. A composite loss function incorporating voxel accuracy, object count consistency, and boundary distance improves segmentation fidelity. The proposed model achieves a Dice score of 95.12 and panoptic quality of 97.01 across microscopy datasets while demonstrating strong generalization to clinical imaging datasets. In addition to the development of these deep learning architectures, this dissertation introduces OrganixInsight, a web-based imaging bioinformatics platform for managing and analyzing high-content microscopy datasets. The platform integrates deep learning models for image restoration and segmentation with tools for experimental design, dataset management, morphometric feature extraction, and interactive 3D visualization. Together, these contributions provide an end-to-end computational framework for high-content screening and phenotypic profiling of organoid models. By combining advanced machine learning techniques with scalable software infrastructure, this work enables automated analysis of large microscopy datasets and provides new opportunities for studying cellular organization, tumor progression, and drug response in complex biological systems."]},{"key":"dc:format","label":"Dc Format","values":["PDF"]},{"key":"dc:title","label":"Title","values":["OrganixInsights: High throughput imaging and high content screening of organoids"]}]}],"canonical_facts":{"dc:contributor.advisor":["Parvin, Bahram"],"dc:contributor.committeemember":["Alvarez Ponce, David","Zhu, Xiaoshan","Xu, Hao","Shen, Yantao","Parvin, Bahram"],"dc:creator":["Sohaib, Muhammad"],"dc:date":["01/01/2026"],"dc:date.accessioned":["2026-06-25T16:22:08Z"],"dc:date.available":["2026-06-25T16:22:08Z"],"dc:date.issued":["2026"],"dc:description.abstract":["Advances in biomedical imaging technologies have significantly expanded the ability of researchers to study complex biological systems at cellular and subcellular resolution. In particular, three-dimensional (3D) organoid models have emerged as powerful experimental systems for investigating tissue development, disease progression, and therapeutic response. Organoids replicate key structural and functional characteristics of in-vivo tissues, including cellular heterogeneity, spatial organization, and extracellular matrix interactions. However, high-resolution imaging of organoids generates large volumetric datasets that require advanced computational methods for analysis. This dissertation presents a comprehensive computational framework for high-content screening and quantitative analysis of 3D organoid and biomedical imaging datasets. The proposed framework integrates deep learning models for image restoration and segmentation with a scalable web-based platform for managing and analyzing microscopy data. First, we introduce Deconv3D, a hybrid transformer–convolution architecture designed for volumetric microscopy restoration. The model integrates convolutional feature extraction with windowed self-attention modules to capture both local spatial features and long-range contextual relationships within volumetric datasets. A novel attention-weighted fusion (AWF) mechanism enables adaptive integration of encoder and decoder features during image reconstruction. Experimental results demonstrate that Deconv3D achieves high-quality reconstruction with a peak signal-to-noise ratio of 30.33 dB and structural similarity index of 0.875, while maintaining a compact architecture with only 2.23 million parameters. Second, we propose 3D-Organoid-SwinNet, a transformer-based architecture for segmentation of nuclei in organoid microscopy datasets. The model combines Swin Transformer encoders with multiscale decoding layers to capture both global contextual information and fine-grained spatial features. Evaluation on breast cancer organoid datasets demonstrates that the model achieves a Dice score of 94.91, outperforming several existing segmentation approaches. Third, we present MAT3D, a multi-aperture transformer architecture for volumetric biomedical image segmentation. MAT3D introduces parallel transformer modules that process multiple spatial representations of the input volume, enabling improved modeling of complex structures. A composite loss function incorporating voxel accuracy, object count consistency, and boundary distance improves segmentation fidelity. The proposed model achieves a Dice score of 95.12 and panoptic quality of 97.01 across microscopy datasets while demonstrating strong generalization to clinical imaging datasets. In addition to the development of these deep learning architectures, this dissertation introduces OrganixInsight, a web-based imaging bioinformatics platform for managing and analyzing high-content microscopy datasets. The platform integrates deep learning models for image restoration and segmentation with tools for experimental design, dataset management, morphometric feature extraction, and interactive 3D visualization. Together, these contributions provide an end-to-end computational framework for high-content screening and phenotypic profiling of organoid models. By combining advanced machine learning techniques with scalable software infrastructure, this work enables automated analysis of large microscopy datasets and provides new opportunities for studying cellular organization, tumor progression, and drug response in complex biological systems."],"dc:format":["PDF"],"dc:identifier.uri":["https://scholarwolf.unr.edu/handle/11714/11889"],"dc:language":["English"],"dc:language.iso":["en_US"],"dc:subject":["3D Organoids","Deep Learning","High-Content Screening","Image Segmentation"],"dc:title":["OrganixInsights: High throughput imaging and high content screening of organoids"],"dc:type":["Dissertation"],"thesis:degree_level":["Doctorate Degree"]},"updated_at":"2026-07-27T21:46:52Z"}