Proceedings of SPIE--the International Society for Optical Engineering最新文献

{"title":"Tumor-SAM: Segment Anything Model for Semi-automatic Lung Tumor Segmentation in CT.","authors":"L Xie, Y Tong, C Wu, D A Torigian, J K Udupa, Y Wan","doi":"10.1117/12.3087226","DOIUrl":"10.1117/12.3087226","url":null,"abstract":"<p><p>Lung tumor segmentation in thoracic CT scans is vital for radiomics analysis and treatment assessment, but is hindered by heterogeneous tumor morphology, ambiguous boundaries, and inaccuracy and labor-intensiveness of manual segmentation. Existing methods, including traditional machine learning and deep learning methods, often suffer from over-/under-segmentation or poor robustness. In this study, we propose an improved Segment Anything Model (called Tumor-SAM) for semi-automatic lung tumor segmentation, integrating U-Net for multi-scale feature extraction and a novel ellipse prompt. Tumor-SAM first detects lung ROI to reduce interference from the surrounding tissue. Then, we design an ellipse prompt defined by center, axes, and rotation that captures tumor shape/location better than points, boxes, or circles. The architecture of Tumor-SAM includes a U-Net-based image encoder (replacing ViT), prompt encoder with positional encoding, multi-head attention fusion, and mask decoder. Our method achieved an average Dice index of 0.84±0.13 and an average Hausdorff distance of 7.25±6.24 mm on 164 testing scans, demonstrating good lung tumor segmentation accuracy.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13929 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13464897/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148727899","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"gDAIS: A Global Analysis Application for Obtaining Confidence in FLIM.","authors":"Nikhil Mattu, Jasmine Su, Siem Yonas, Yuyi Liu, Greg Alspaugh, Shureed Qazi, Ayman Slamani, Katie A Link, Rozhin Penjweini, Christian A Combs, Dan Sackett, Jay R Knutson","doi":"10.1117/12.3081144","DOIUrl":"10.1117/12.3081144","url":null,"abstract":"<p><p>Fluorescence Lifetime Imaging Microscopy (FLIM) is a popular imaging technique that provides users another dimension for investigating biomolecular states, interactions, and environments. It can, for instance, be used to monitor metabolism or discern different Förster resonance energy transfer (FRET) proximity relationships within cells. FLIM often reveals multiple lifetimes which are usually analyzed via multi-exponential fitting in the time domain and phasor localization in the frequency domain. Both methods yield maps of amplitudes (i.e., alphas/intensities/concentrations) associated with their lifetimes. The resulting \"pixel histograms\" or \"phasor clouds\" give the user pixel-weighted visualizations of the distribution of amplitudes vs. lifetime. Using this approach, however, pixel areas with low photon counts can improperly skew FLIM analysis. In contrast, a global decay-associated analysis of image data first optimizes the nonlinear parameters that are common across the image-namely, the lifetimes-along with their amplitudes, and then provides statistically rigorous confidence limits on both lifetimes and amplitudes. With these limits, one may assess cellular metabolism and FRET relationships with confidence. Additionally, global analysis is known to be capable of unmixing multiple components that are inseparable in conventional or graphical analyses. We present global Decay-Associated Image Software (gDAIS) for analyzing FLIM data which incorporates a fast regression method and yields images of the amplitudes and images of all associated uncertainties. These outputs can then be used to generate error limits for amplitude ratios (e.g., free:bound NADH).</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13856 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13007990/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147517281","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Toward a Surgical Clip-Based Registration Approach for Aligning Pathology Data to Tumor Bed Using Mixed Reality.","authors":"Richie Dix, Mays Neiroukh, Bowen Xiang, Jie Ying Wu, Michael C Topf, Ingrid M Meszoely, Jon S Heiselman, Michael I Miga","doi":"10.1117/12.3087843","DOIUrl":"10.1117/12.3087843","url":null,"abstract":"&lt;p&gt;&lt;p&gt;During tumor resections, surgeons must integrate visual cues from the operative field with feedback from pathology to evaluate resection margins. This process requires constructing a complex mental spatial map of the tumor bed and the ability to integrate pathological analysis back for additional resections. Unfortunately, this can lead to incomplete resections and result in positive margins. Such outcomes often necessitate additional surgeries, increasing patient risk, anxiety, and overall healthcare burden. To address this challenge, we are investigating a mixed reality (MR) system that transforms pathology-confirmed positive margins directly to locations within the surgical field. A critical requirement of this system is accurate alignment of pathology data with the intraoperative presentation of the tumor bed. We present and evaluate a registration workflow that employs surgical clips placed on both resection specimens and corresponding tumor bed sites as corresponding fiducial markers. Data acquisition was performed using the Microsoft HoloLens&lt;sup&gt;®&lt;/sup&gt; 2 enabled to track a conventional stylus fitted with conventional retroreflective spheres. Voice commands were incorporated to streamline point collection and registration. System point acquisition accuracy and precision were assessed using a machined phantom containing nine fiducial points, each with a known ground truth and positioned throughout the field of view. Precision was measured from the point distribution of repetitive acquisition, while accuracy was evaluated by registering collected data to ground-truth coordinates. To investigate our MR approach within the desired clinical context, a second soft tissue-like breast phantom was created to simulate a mock resection. Conventional surgical clips used in surgical practice were used as fiducial markers, and target points within the mock resection site were created using a process that ensured known correspondence between the mock specimen and the tumor-bed. It should be noted that in this preliminary realization, rigid alignment approaches were used. Ground truth comparisons were independently captured using a second optical tracker. Following rigid registration, specimen targets were projected into the phantom cavity using the HoloLens&lt;sup&gt;®&lt;/sup&gt; 2 and compared against the ground truth. The point acquisition precision yielded a fiducial localization error (FLE) of 0.7 ± 0.2 mm. When registering to the ground-truth machined phantom, the fiducial registration error (FRE) was 1.8 ± 0.6 mm and the target registration error (TRE) was 2.6 ± 0.9 mm. With respect to the mock clinical scenario, the deep-margin gold-standard alignment error using the conventional optical tracker yielded an FRE of 2.8 ± 1.0 mm with a mean TRE of 3.3 ± 1.7 mm. When performing the same experiment using HoloLens&lt;sup&gt;®&lt;/sup&gt; 2 internal tracking, the composite mean FRE increased to 3.8 ± 0.8 mm and the composite mean TRE increased to 5.3 ± 1.1mm. Overall, t","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13927 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13479359/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148802485","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Stereo Cherenkov imaging as 3D absolute dose monitoring tool for total skin electron therapy (TSET).","authors":"Yifeng Zhu, Baozhu Lu, Emily Xiong, Ching Lai, Karen Tang, Harper G Hubbeling, John P Plastaras, Michael LaRiviere, Brian W Pogue, Timothy C Zhu","doi":"10.1117/12.3090142","DOIUrl":"https://doi.org/10.1117/12.3090142","url":null,"abstract":"<p><p>Cherenkov imaging provides a non-contact method for monitoring surface dose delivery during total skin electron therapy (TSET), where patients are treated in six postures using the Stanford standing technique. In this study, a multi-view Cherenkov imaging framework combined with patient-specific correction factors was investigated to improve quantitative three-dimensional (3D) surface dose estimation. Cherenkov emission was recorded during treatment using three synchronized time-gated cameras providing anterior and lateral coverage. Patient-specific anatomical models reconstructed from 3D surface scans were used to generate Monte Carlo (MC) reference dose distributions. Cherenkov images were corrected for background, camera response, and geometric distortions before conversion to pseudo-dose using in-vivo dosimetry calibration. Vertical Cherenkov intensity profiles were compared with MC dose profiles to derive patient-specific correction factors. Results showed that uncorrected Cherenkov-derived dose underestimated dose in distal and lateral body regions, while applying patient-specific corrections and multi-view imaging improved agreement with MC reference dose, supporting enhanced surface dose monitoring for TSET.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13841 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13638229/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149119721","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implementation of a Prospective, Multi-Center Clinical Study of Dynamic Contrast Fluorescence Imaging to Diagnose Necrotizing Soft-Tissue Infection - Challenges and Lessons Learned.","authors":"Eric R Henderson, Morgan T Mazanec, Chelsey L Recendez, Jon M Anderson, Cameron Woods, Logan M Bateman, Zachary Soucy, Scott W Rodi, Ida Leah Gitajn, Jaimo Ahn, Sunil Singhal, Niels D Martin, Jane Keating, Nicholas M Bernthal, Christopher Lee, Eben Rosenthal, William T Obremskey, Jonathan G Schoenecker, Joseph Forrester, Kurt Weiss, Graciela Bauza, Lauren Rosenblum, Rebecca E Kotcher, Mary Molly Hunter, Raymond Jean, Samuel S Streeter, Jonathan T Elliott","doi":"10.1117/12.3093146","DOIUrl":"10.1117/12.3093146","url":null,"abstract":"<p><p>Necrotizing soft-tissue infections (NSTIs) are rapidly progressive and highly lethal, with diagnostic uncertainty contributing to treatment delays and mortality rates approaching 30%. To address this challenge, a multi-center clinical trial has been initiated to evaluate dynamic contrast-enhanced fluorescence imaging (DCE-FI) with indocyanine green as a rapid diagnostic tool. This technology enables visualization of microvascular thrombi characteristic of NSTIs, distinguishing them from less severe infections. The trial engages seven tertiary centers within a collaborative NSTI study group. This presentation will focus on the execution and implementation of this complex, multi-institutional study, including processes for protocol harmonization, regulatory coordination, and standardized imaging workflows. A key element of implementation was the development of real-time, smartphone-accessible didactic materials to ensure consistent imaging technique, data acquisition, and troubleshooting across all participating sites. Emphasis will be placed on the strategies used to maintain study fidelity and ensure successful completion across centers, enabling validation of DCE-FI as a transformative diagnostic adjunct in NSTI care.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13832 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13042105/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147610895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Computational Pipeline for Patient-Specific Modeling of Thoracic Aortic Aneurysm: From Medical Image to Finite Element Analysis.","authors":"Jiasong Chen, Linchen Qian, Ruonan Gong, Christina Sun, Tongran Qin, Thuy Pham, Caitlin Martin, Mohammad Zafar, John Elefteriades, Wei Sun, Liang Liang","doi":"10.1117/12.3087786","DOIUrl":"10.1117/12.3087786","url":null,"abstract":"<p><p>Thoracic aortic aneurysm (TAA) represents a critical cardiovascular challenge, characterized by the silent, progressive dilation of the aorta that can lead to catastrophic rupture or dissection. Despite the significant clinical burden of TAA, current clinical risk stratification relies heavily on maximum diameter measurements derived from three-dimensional computed tomography (3D CT), a metric that often fails to capture the complex biomechanical environment precipitating failure. To address this, we developed a comprehensive computational pipeline for patient-specific modeling of thoracic aortic aneurysms, combining segmentation, mesh generation, and finite element simulation. A key innovation of this pipeline is the open-arch mesh with node and element correspondence among different patients, enabled by the template-fitting based meshing method, which preserves anatomical accuracy, enables statistical shape modeling, and facilitates FE simulations with all-hexahedral elements in PyTorch-FEA -capabilities not addressed in previous studies. Our pipeline captures clinically relevant biomechanical differences, showing that aneurysmal patients exhibit significantly higher maximum absolute principal stress compared to non-aneurysmal patients. This result identifies a quantitative biomechanical marker that extends beyond conventional geometric measurements and potentially aids in TAA diagnosis. Moreover, the pipeline quantifies stress distribution across different regions of the aorta, enabling more detailed biomechanical assessment and supporting personalized treatment planning. By streamlining the transition from routine imaging to advanced stress analysis, this pipeline facilitates large-scale population studies and paves the way for the integration of patient-specific biomechanical parameters into routine clinical decision-making.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13929 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13576376/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148979132","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Organoid Tracker: A SAM2-Powered Platform for Zero-shot Cyst Analysis in Human Kidney Organoid Videos.","authors":"Xiaoyu Huang, Lauren M Maxson, Trang Nguyen, Cheng Jack Song, Yuankai Huo","doi":"10.1117/12.3086163","DOIUrl":"10.1117/12.3086163","url":null,"abstract":"<p><p>Recent advances in organoid models have revolutionized the study of human kidney disease mechanisms and drug discovery by enabling scalable, cost-effective research without the need for animal sacrifice. Here, we present a kidney organoid platform optimized for efficient screening in polycystic kidney disease (PKD). While these systems generate rich spatial-temporal microscopy video datasets, current manual approaches to analysis remain limited to coarse classifications (e.g., hit vs. non-hit), often missing valuable pixel-level and longitudinal information. To help overcome this bottleneck, we develop Organoid Tracker, a graphical user interface (GUI) platform designed with a modular plugin architecture, which empowers researchers to extract detailed, quantitative metrics without programming expertise. Built on the cutting-edge vision foundation model Segment Anything Model 2 (SAM2), Organoid Tracker enables zero-shot segmentation and automated analysis of spatial-temporal microscopy videos. It quantifies key metrics such as cyst formation rate, growth velocity, and morphological changes, while generating comprehensive reports. By providing an extensible, open-source framework, Organoid Tracker offers a powerful solution for improving and accelerating research in kidney development, PKD modeling, and therapeutic discovery. The platform is publicly available as open-source software at https://github.com/hrlblab/OrganoidTracker.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13930 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13637688/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149119791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CAVASS-DL: An Enhanced Computer-Assisted Visualization and Analysis Software System by Integrating Deep Learning Segmentation Models.","authors":"L Xie, G J Grevera, Y Tong, C Wu, Z Liu, D A Torigian, J K Udupa","doi":"10.1117/12.3087224","DOIUrl":"10.1117/12.3087224","url":null,"abstract":"<p><p>Segmentation in medical images plays a crucial first step for the quantitative analysis of organs of interest and disease conditions that may affect them. In the 2000s, we released an open-source medical image processing and analysis software system called CAVASS (Computer-Assisted Visualization and Analysis Software System), whose lineage goes back to 1980 as the first software for medical image processing. CAVASS supports multiple types of medical images and numerous usage scenarios/use cases. To allow ready access to advanced deep learning models within the easy-to-use graphical user interface (GUI) of CAVASS to greatly facilitate ground truth segmentation generation and production mode image segmentation, we created an enhanced version of CAVASS, called CAVASS-DL, by integrating the Segment Anything Model (SAM) within our software, which can extract and fuse the features from images and interactive prompts to enhance the segmentation performance. Firstly, we modified the GUI to enable intuitive region-of-interest (ROI) prompting via bounding boxes. Secondly, we implemented and trained the SAM network using PyTorch platform. Lastly, we bundled the SAM network as a standalone executable interface for CAVASS. The illustrative and quantitative results on 20 dynamic MRI scans of neck and thorax demonstrate that our approach achieves high accuracy and efficiency for semi-automatic anatomy object segmentation via CAVASS-DL in a convenient manner.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13927 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13676771/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149221288","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Optical Coherence Tomography Harmonization via Dual Diffusion Implicit Bridges.","authors":"Shuwen Wei, Samuel W Remedios, Blake E Dewey, Zhangxing Bian, Shimeng Wang, Junyu Chen, Yihao Liu, Bruno Jedynak, Tin Y A Liu, Shiv Saidha, Peter A Calabresi, Jerry L Prince, Aaron Carass","doi":"10.1117/12.3086131","DOIUrl":"10.1117/12.3086131","url":null,"abstract":"<p><p>Image harmonization (IH) makes images from different domains consistent, enabling multi-domain comparison with reliable quantitative measurements. Optical coherence tomography (OCT) could benefit from IH, as variations in OCT systems cause significant discrepancies in image quality due to speckle and noise. However, IH approaches typically rely on paired images during the training phase, and such images are not available in OCT data due to sparse image acquisition. A Schrödinger bridge (SB) finds an optimal coupling between arbitrary probability spaces with respect to a reference path measure and can be applied to unpaired IH. SBs are the core of many generative diffusion models, when one of the probability spaces is easy to sample from (i.e., Gaussian). SBs have been applied to natural images for image-to-image translation; however, they have not been applied to OCT IH because a SB does not guarantee anatomical consistency. In this paper, we use a dual diffusion implicit bridge (DDIB) for OCT IH, which finds independent SBs between arbitrary domains and a common Gaussian probability space. We generalize the assumption of DDIB and use segmentation masks as a condition to improve anatomical consistency. We conduct DDIB experiments with and without segmentation masks as a condition and analyze performance in terms of anatomical consistency and harmonization quality. The method does not require any paired training data and can in principle be quickly adapted to new domains.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13925 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13352811/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148426658","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"12-Month Disease-Free Survival Prediction in Diffuse Large B-Cell Lymphoma: A Multimodal Transfer Learning Innovation.","authors":"Xinwei Wang, Tiange Liu, Jian Dai, Emeline Chong, Stephen J Schuster, Drew A Torigian, Yubing Tong, Jayaram K Udupa","doi":"10.1117/12.3086292","DOIUrl":"https://doi.org/10.1117/12.3086292","url":null,"abstract":"<p><p>Early relapse of diffuse large B-cell lymphoma (DLBCL) within 12 months of diagnosis confers a poor prognosis, underscoring the critical need for accurate pretreatment prediction. Current approaches relying on radiomics and clinical indicators are limited by scarce medical imaging data, time-consuming manual feature extraction (susceptible to human subjectivity), and insufficient use of multimodal deep learning frameworks. To address these challenges, we propose MM-DLBCL, a novel multimodal transfer learning framework for pretreatment 12-month disease-free survival (DFS12) prediction. It leverages the biomedical pre-trained PMC-CLIP model to mitigate small-sample constraints, uses pretreatment FDG-PET/CT scans to generate coronal and sagittal Maximum Intensity Projection (MIP) images (focusing on metabolically active lesions), and converts key clinical covariates (e.g., International Prognostic Index (IPI) score, disease stage, serum lactate dehydrogenase (LDH) levels) into structured text embeddings. The architecture integrates an enhanced ResNet50 visual encoder (for dual-plane MIP feature extraction), a PubMedBERT text encoder (for clinical data processing), a Medical-Statistical Feature Fusion (MSFF) module (to quantify tumor heterogeneity via higher-order statistics), a Multi-Scaled Cooperative Attention (MSCA) module (to align imaging and clinical features), and a Cross-Modal Contrastive Head (CMCH) module (to optimize discriminative power via triple contrastive loss). Trained and validated on a retrospective cohort of 338 DLBCL adult patients, MM-DLBCL achieved an end-to-end AUC of 0.8316 ± 0.06 and accuracy of 0.8044 ± 0.05 for DFS12 prediction, demonstrating significant potential for multimodal transfer learning in automated DLBCL disease-free survival risk stratification.</p>","PeriodicalId":74505,"journal":{"name":"Proceedings of SPIE--the International Society for Optical Engineering","volume":"13925 ","pages":""},"PeriodicalIF":0.0,"publicationDate":"2026-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13681197/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149240747","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}