{"id":{"repo_id":"bu","oai_identifier":"oai:open.bu.edu:2144/52121"},"canonical_url":"https://search.ndltd.org/etd/bu/oai:open.bu.edu:2144/52121","repository":{"repo_id":"bu","name":"Boston University","base_url":"https://open.bu.edu/oai/request"},"display":{"title":"Advancing computational multi-aperture fluorescence microscopy via model- and learning-based techniques","abstract":"Fluorescence microscopy is indispensable for studying biological structure and dynamics, yet emerging applications demand micron-scale resolution over centimeter fields of view (FOV) in a compact, portable form factor. Recent miniaturized lensless designs replace objectives with a phase mask to achieve extreme compactness, but the resulting strong multiplexing can reduce measurement contrast and signal-to-noise ratio (SNR), limiting robustness for weak fluorescence and in the presence of tissue scattering. Computational Miniature Mesoscope (CM2) takes an alternative approach by placing a microlens array (MLA) in front of a single CMOS sensor: each microlens images a sub-FOV, and controlled overlap maximizes usable FOV, encoding extended high-resolution scenes on a compact device. This optical simplicity shifts complexity to computation: the periodic MLA makes the inverse problem ill-conditioned and produces highly non-local, spatially varying point spread functions (PSFs). This dissertation develops a unified modeling-and-learning framework that explicitly accounts for this spatial variance and enables uniform, wide-FOV reconstructions from CM2 measurements. I first formulate a low-rank linear shift-variant (LSV) forward model by factorizing sparse 3D PSFs with truncated SVD (TSVD) and representing field dependence with smoothly varying coefficient maps. Using this model, I derive an LSV–ADMM reconstruction algorithm that improves peripheral fidelity and overall image quality relative to shift-invariant baselines, and I use the calibrated LSV simulator to generate diverse, high-fidelity training data. Building on this foundation, I introduce three learning-based reconstructions tailored to CM2: CM2Net, which combines view demixing, synthesis, and refocusing to enable single-shot high-resolution 3D particle localization; SVFourierNet, which performs global low-rank deconvolution in the Fourier domain with learned inverse filters for fast spatially variant reconstruction across a wide field; and SV-CoDe, which modulates shared convolutional features with coordinate-driven coefficients predicted by a lightweight multi-layer perceptron (MLP), improving peripheral fidelity while decoupling parameter and memory costs from FOV size. Together, these contributions deliver a practical end-to-end solution that combines a calibrated low-rank spatially varying forward model, a model-based optimization baseline, and scalable spatially varying deep networks to address MLA multiplexing and spatially varying blur, enabling high-throughput, uniform wide-FOV fluorescence imaging on compact hardware.","abstract_html":"Fluorescence microscopy is indispensable for studying biological structure and dynamics, yet emerging applications demand micron-scale resolution over centimeter fields of view (FOV) in a compact, portable form factor. Recent miniaturized lensless designs replace objectives with a phase mask to achieve extreme compactness, but the resulting strong multiplexing can reduce measurement contrast and signal-to-noise ratio (SNR), limiting robustness for weak fluorescence and in the presence of tissue scattering. Computational Miniature Mesoscope (CM2) takes an alternative approach by placing a microlens array (MLA) in front of a single CMOS sensor: each microlens images a sub-FOV, and controlled overlap maximizes usable FOV, encoding extended high-resolution scenes on a compact device. This optical simplicity shifts complexity to computation: the periodic MLA makes the inverse problem ill-conditioned and produces highly non-local, spatially varying point spread functions (PSFs). This dissertation develops a unified modeling-and-learning framework that explicitly accounts for this spatial variance and enables uniform, wide-FOV reconstructions from CM2 measurements. I first formulate a low-rank linear shift-variant (LSV) forward model by factorizing sparse 3D PSFs with truncated SVD (TSVD) and representing field dependence with smoothly varying coefficient maps. Using this model, I derive an LSV–ADMM reconstruction algorithm that improves peripheral fidelity and overall image quality relative to shift-invariant baselines, and I use the calibrated LSV simulator to generate diverse, high-fidelity training data. Building on this foundation, I introduce three learning-based reconstructions tailored to CM2: CM2Net, which combines view demixing, synthesis, and refocusing to enable single-shot high-resolution 3D particle localization; SVFourierNet, which performs global low-rank deconvolution in the Fourier domain with learned inverse filters for fast spatially variant reconstruction across a wide field; and SV-CoDe, which modulates shared convolutional features with coordinate-driven coefficients predicted by a lightweight multi-layer perceptron (MLP), improving peripheral fidelity while decoupling parameter and memory costs from FOV size. Together, these contributions deliver a practical end-to-end solution that combines a calibrated low-rank spatially varying forward model, a model-based optimization baseline, and scalable spatially varying deep networks to address MLA multiplexing and spatially varying blur, enabling high-throughput, uniform wide-FOV fluorescence imaging on compact hardware.","abstract_has_math":false,"creators":["Yang, Qianwan"],"institution":"Boston University","degree_name":"Doctor of Philosophy","degree_level":"doctoral","degree_discipline":"Electrical & Computer Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Tian, Lei"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-19T02:42:12Z","subjects":["Electrical engineering","3D particle localization","Computational imaging","Multi-aperture imaging","Physics-informed learning","Spatially varying deconvolution"],"languages":["en_US"],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[{"key":"dc:identifier.orcid","label":"ORCID","values":["0000-0002-5504-4970"],"render_values":[{"text":"0000-0002-5504-4970","href":"https://orcid.org/0000-0002-5504-4970","code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/2144/52121","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"https://hdl.handle.net/2144/52121","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://open.bu.edu/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Aopen.bu.edu%3A2144%2F52121","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Tian, Lei"]},{"key":"dc:creator","label":"Author","values":["Yang, Qianwan"]},{"key":"dc:identifier.orcid","label":"ORCID","values":["0000-0002-5504-4970"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-01-30T15:37:45Z"]},{"key":"dc:date.embargo","label":"Dc Date Embargo","values":["2028-01-29T00:00:00Z"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:date.updated","label":"Dc Date Updated","values":["2026-01-30T02:02:20Z"]},{"key":"dc:type","label":"Dc Type","values":["Thesis/Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical & Computer Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Boston University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Electrical engineering","3D particle localization","Computational imaging","Multi-aperture imaging","Physics-informed learning","Spatially varying deconvolution"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en_US"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/2144/52121"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["2026"]},{"key":"dc:description.abstract","label":"Abstract","values":["Fluorescence microscopy is indispensable for studying biological structure and dynamics, yet emerging applications demand micron-scale resolution over centimeter fields of view (FOV) in a compact, portable form factor. Recent miniaturized lensless designs replace objectives with a phase mask to achieve extreme compactness, but the resulting strong multiplexing can reduce measurement contrast and signal-to-noise ratio (SNR), limiting robustness for weak fluorescence and in the presence of tissue scattering. Computational Miniature Mesoscope (CM2) takes an alternative approach by placing a microlens array (MLA) in front of a single CMOS sensor: each microlens images a sub-FOV, and controlled overlap maximizes usable FOV, encoding extended high-resolution scenes on a compact device. This optical simplicity shifts complexity to computation: the periodic MLA makes the inverse problem ill-conditioned and produces highly non-local, spatially varying point spread functions (PSFs). This dissertation develops a unified modeling-and-learning framework that explicitly accounts for this spatial variance and enables uniform, wide-FOV reconstructions from CM2 measurements. I first formulate a low-rank linear shift-variant (LSV) forward model by factorizing sparse 3D PSFs with truncated SVD (TSVD) and representing field dependence with smoothly varying coefficient maps. Using this model, I derive an LSV–ADMM reconstruction algorithm that improves peripheral fidelity and overall image quality relative to shift-invariant baselines, and I use the calibrated LSV simulator to generate diverse, high-fidelity training data. Building on this foundation, I introduce three learning-based reconstructions tailored to CM2: CM2Net, which combines view demixing, synthesis, and refocusing to enable single-shot high-resolution 3D particle localization; SVFourierNet, which performs global low-rank deconvolution in the Fourier domain with learned inverse filters for fast spatially variant reconstruction across a wide field; and SV-CoDe, which modulates shared convolutional features with coordinate-driven coefficients predicted by a lightweight multi-layer perceptron (MLP), improving peripheral fidelity while decoupling parameter and memory costs from FOV size. Together, these contributions deliver a practical end-to-end solution that combines a calibrated low-rank spatially varying forward model, a model-based optimization baseline, and scalable spatially varying deep networks to address MLA multiplexing and spatially varying blur, enabling high-throughput, uniform wide-FOV fluorescence imaging on compact hardware."]},{"key":"dc:title","label":"Title","values":["Advancing computational multi-aperture fluorescence microscopy via model- and learning-based techniques"]}]}],"canonical_facts":{"dc:contributor.advisor":["Tian, Lei"],"dc:creator":["Yang, Qianwan"],"dc:date.accessioned":["2026-01-30T15:37:45Z"],"dc:date.embargo":["2028-01-29T00:00:00Z"],"dc:date.issued":["2026"],"dc:date.updated":["2026-01-30T02:02:20Z"],"dc:description":["2026"],"dc:description.abstract":["Fluorescence microscopy is indispensable for studying biological structure and dynamics, yet emerging applications demand micron-scale resolution over centimeter fields of view (FOV) in a compact, portable form factor. Recent miniaturized lensless designs replace objectives with a phase mask to achieve extreme compactness, but the resulting strong multiplexing can reduce measurement contrast and signal-to-noise ratio (SNR), limiting robustness for weak fluorescence and in the presence of tissue scattering. Computational Miniature Mesoscope (CM2) takes an alternative approach by placing a microlens array (MLA) in front of a single CMOS sensor: each microlens images a sub-FOV, and controlled overlap maximizes usable FOV, encoding extended high-resolution scenes on a compact device. This optical simplicity shifts complexity to computation: the periodic MLA makes the inverse problem ill-conditioned and produces highly non-local, spatially varying point spread functions (PSFs). This dissertation develops a unified modeling-and-learning framework that explicitly accounts for this spatial variance and enables uniform, wide-FOV reconstructions from CM2 measurements. I first formulate a low-rank linear shift-variant (LSV) forward model by factorizing sparse 3D PSFs with truncated SVD (TSVD) and representing field dependence with smoothly varying coefficient maps. Using this model, I derive an LSV–ADMM reconstruction algorithm that improves peripheral fidelity and overall image quality relative to shift-invariant baselines, and I use the calibrated LSV simulator to generate diverse, high-fidelity training data. Building on this foundation, I introduce three learning-based reconstructions tailored to CM2: CM2Net, which combines view demixing, synthesis, and refocusing to enable single-shot high-resolution 3D particle localization; SVFourierNet, which performs global low-rank deconvolution in the Fourier domain with learned inverse filters for fast spatially variant reconstruction across a wide field; and SV-CoDe, which modulates shared convolutional features with coordinate-driven coefficients predicted by a lightweight multi-layer perceptron (MLP), improving peripheral fidelity while decoupling parameter and memory costs from FOV size. Together, these contributions deliver a practical end-to-end solution that combines a calibrated low-rank spatially varying forward model, a model-based optimization baseline, and scalable spatially varying deep networks to address MLA multiplexing and spatially varying blur, enabling high-throughput, uniform wide-FOV fluorescence imaging on compact hardware."],"dc:identifier.orcid":["0000-0002-5504-4970"],"dc:identifier.uri":["https://hdl.handle.net/2144/52121"],"dc:language.iso":["en_US"],"dc:subject":["Electrical engineering","3D particle localization","Computational imaging","Multi-aperture imaging","Physics-informed learning","Spatially varying deconvolution"],"dc:title":["Advancing computational multi-aperture fluorescence microscopy via model- and learning-based techniques"],"dc:type":["Thesis/Dissertation"],"thesis:degree_discipline":["Electrical & Computer Engineering"],"thesis:degree_level":["doctoral"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Boston University"]},"updated_at":"2026-11-19T02:42:12Z"}