UCLA Technology Development Group的动态

Integrating features in medical imaging-based classification problems, such as those involving magnetic resonance imaging-based diagnosis, remains a significant challenge. Current approaches to multimodal medical imaging classification often focus on features from a single modality or concatenate them into a high-dimensional feature vector, resulting in overfitting. This paper presents a novel method in deep fusion based on depthwise 1D convolution. The effectiveness of this method is demonstrated through experiments involving the brain tumor segmentation (BraTs-21) competition, Task-2, which involves predicting O(6)-methylguanine-DNA methyltransferase promoter status, a biomarker for glioblastoma, as a binary classification problem using multimodal magnetic resonance images. The experimental results show the promising performance of this approach. The proposed method got the best AUC score of 0.748 with minimum BCE loss of 0.62 on BraTs-21 as compared to other methods.

My Edited book "Applications of Artificial Intelligence in Medical Imaging" has recently been published by Elsevier. Applications of Artificial Intelligence in Medical Imaging?describes various biomedical image analysis in disease detection using AI that can be used to incorporate knowledge obtained from different medical imaging devices such as CT, X-ray, PET and ultrasound. The book discusses the use of AI for the detection of several cancer types, including brain tumor, breast, pancreatic, rectal, lung colon, and skin. In addition, it explains how AI and deep learning techniques can be used to diagnose Alzheimer's, Parkinson's, COVID-19 and mental conditions. This is a valuable resource for clinicians, researchers and healthcare professionals who are interested in learning more about AI and its impact in medical/biomedical image analysis. You can find the details in: https://lnkd.in/ezw7v3ep

I'm thrilled to announce that my latest research paper on melanoma detection has been published in SAGE Journals! (Farrukh A. Khan, PhD, FBCS, SMIEEE, Mamoona HUmayun) Discover how our approach enhances privacy and accuracy in medical imaging. Read it here: https://lnkd.in/eckYtskQ I look forward to your thoughts and feedback! #Research #MelanomaDetection #HyperParameterTuning #CNN #MedicalImaging #AcademicPublishing

?? New Publication Alert in Medical Physics! ?? "Deep learning (#DL) based optimization of field geometry for total marrow irradiation (#TMI) delivered with volumetric modulated arc therapy" For the first time, an AI/DL approach has been successfully applied to enhance the precision of TMI isocenters positioning—a vital aspect of treatment planning that directly impacts the feasibility of implementing this technique in non-dedicated centers. Our study represents a pioneering effort to leverage AI in improving TMI procedures, significantly reducing the toxicities in patients slated for bone marrow transplantation. https://lnkd.in/dnBudDeU ?? Open Access Tool We’ve made our innovative tool freely available for download. It integrates seamlessly into treatment planning systems, enabling medical physicists worldwide to refine their strategies with AI-driven insights. Get your hands on this tool. https://lnkd.in/dnBudDeU ?? Acknowledgments This achievement wouldn't have been possible without the dedicated efforts of our PhD teams (Nicola Lambri, Damiano Dei, #RicardoCoimbraBrioso Leonardo Crespi), who have worked to push the boundaries of what's possible in medical treatment planning. A special thanks to Giorgio Longari, whose master's thesis laid the foundation for this groundbreaking research. ?? Let's Collaborate Are you involved in healthcare or AI research? We’re on the lookout for new partners to test and collaborate on this exciting technology. Write to us if you're interested in exploring how this groundbreaking tool can be applied in your practice or research. ?? Thank You This study was part of the #AuToMI project, funded by the Italian Ministry of Health. We extend our heartfelt thanks for their support in advancing medical technology. Daniele Loiacono, Carmela Galdieri, Francesca Lobefalo, Giacomo Reggiori, Roberto Rusconi, Stefano Tomatis, #LuisaBellu, #StefaniaBramanti, #ElenaClerici, chiara de philippis, piera navarria, Carmelo Carlo-Stella, Ciro Franzese, Marta Scorsetti #MedicalPhysics #AIinHealthcare #Innovation #RadiationTherapy #Collaboration #HealthTech

"Biomedical Engineering Roadmap with a Focus on Artificial Intelligence": This article delves into the challenges and opportunities that lie ahead in the field of biomedical engineering, with a particular focus on innovations driven by artificial intelligence. The paper discusses how AI can enhance diagnostic and therapeutic processes, outlining the future landscape of biomedical engineering and the role AI is expected to play in advancing this field. https://lnkd.in/d-HrV7nQ #DeepLearning #MedicalImaging #MultiOrganSegmentation #ArtificialIntelligence #AIinHealthcare #SupervisedLearning #WeaklySupervisedLearning #SemiSupervisedLearning #BiomedicalEngineering #HealthTech #ImageAnalysis #AIResearch

We’re excited to work with leading biomedical data analysis platform provider Velsera and the Center for Data-Driven Discovery in Biomedicine (D3b) at Children's Hospital of Philadelphia to advance the analysis of large-scale genomics and medical imaging datasets. Read all about how these two medical data management and analysis platforms come together to enable better genomics variant identification and provide enhanced prediction of clinical outcomes for individuals with Down syndrome. #imaging #medcialimaging #imagingai #downsyndromeresearch #AI

"The existence of multiple frameworks catering to radiomics researchers and educators instills optimism for expediting the translation of radiomics into clinical practice." #EuropeanRadiology #ArtificialIntelligence #Radiomics ?? Click the link below to view the latest post on the #AI blog from Hyun Soo Ko and Kevin Tran!

In medical imaging, we often see two separate worlds: academic and industrial. Benchmark challenges comparing algorithms are prevalent in the academic world, but it often needs to be clarified how challenge performance translates to clinical practice. In contrast, algorithms originating from industry must be FDA/CE-certified for clinical use, and the associated validation studies are published. However, they are not compared to other algorithms (industrial or academic). To bridge this gap, in our recent publication in Modern Pathology (https://lnkd.in/efzwfkpe), spearheaded by Khrystyna Faryna, we compared top-performing academic algorithms from the PANDA Challenge (https://lnkd.in/erP4rkiY) to clinical-grade commercial algorithms from Paige and AIRA Matrix Private Limited (big compliments to them for being brave enough to participate!). To assess generalization performance, we crowdsourced 113 biopsies from seven sites and five scanners. None of the algorithms were re-trained on this dataset, and we independently performed the final validation. If you are interested in what came out, I highly recommend reading the paper, and if you want to try out your own algorithm on this dataset, you can access it through https://lnkd.in/eqJ_PZkn #computationalpathology #medicalai #machinelearning

Stable Diffusion Segmentation for Biomedical Images with Single-step Reverse Process ?? In the evolving field of medical image segmentation, a new model called SDSeg has been introduced, representing a major technological advancement. This paper addresses key challenges in the application of diffusion models to medical image segmentation, focusing on resource efficiency and the stability of predictions. Article Link: https://lnkd.in/gM9zWivc ?? Context and Problem Statement: Traditional diffusion models, like Diffusion Probabilistic Models (DPMs), typically involve generating segmentation maps in the pixel space, requiring multiple reverse steps to produce detailed images. This approach is computationally expensive and time-consuming, often needing numerous samples to ensure stable predictions. The process consumes significant computational resources and also poses challenges in maintaining consistency across predictions, which is critical in medical diagnostics. ?? Novel Methodology and Contributions: The paper introduces SDSeg, a latent diffusion segmentation model built on the Stable Diffusion (SD) framework. SDSeg innovates by operating in a perceptually equivalent latent space rather than the full-resolution pixel space. This allows the model to maintain computational efficiency while performing the diffusion process. SDSeg follows a single-step reverse process, enabled by a simple latent estimation loss function. This function helps the model predict segmentation outcomes directly from the latent representation, bypassing the need for multiple sampling steps. SDSeg also uses a latent fusion technique, which integrates the latent representations of the input image and the segmentation map, enhancing the model's ability to capture structural and spatial features necessary for accurate segmentation. ?? Key Findings: The study showcases SDSeg's superior performance on five benchmark datasets, including modalities such as RGB and CT images. It outperforms existing models in terms of the accuracy of segmentation. The model's ability to produce consistent and accurate segmentation with just one reverse step and without multiple samples is a breakthrough, significantly reducing the inference time and computational load. ?? Conclusion and Future Directions: SDSeg sets a new standard in diffusion-based medical image segmentation by combining efficiency with high accuracy. The model's design and its use of a trainable vision encoder and latent space operations, enable it to adapt to various medical imaging modalities. Future work may explore the integration of SDSeg into clinical workflows, potentially enhancing real-time diagnostic capabilities and broadening its applicability to other areas of medical imaging. Lin, T., Zhu, S., Liu, H., Zhang, Q., & Tang, J. (2024). "Stable Diffusion Segmentation for Biomedical Images with Single-step Reverse Process". ArXiv. /abs/2406.18361

Our research group, Computer-assisted Applications in Medicine (CAiM) at Uppsala University, has a new Ph.D. position opening on 'Machine Learning & Image Processing for Radiotherapy.' This is a collaborative project between Uppsala University and Uppsala University Hospital. The latest generation treatment machines (MR-Linac) where an MR camera is integrated, thus enable high-quality imaging of the soft-tissue at the treatment. The precision of the treatment is increased since the treatment is adapted to the daily patient anatomy for a more individiualised treatment. To monitor and adapt treatments given changing and moving anatomy, dose accumulation between multiple treatment sessions is needed. This requires Deformable Image Registration (DIR) for precisely aligning multiple images, but todays’ algorithms are relatively slow, sub-optimal, and unaware of potential errors and uncertainties. A main goal of this project is to develop AI based DIR techniques that are not only faster and more accurate, but also are interpretable and uncertainty aware. The project results have great potential of being used in the clinics and the future of radiotherapy. The application deadline is 28th of February! For more information, visit: https://lnkd.in/dsBw-mtr