AIIIMA’s Post

"Causality matters in medical imaging" joint publication by CHAI Hub co-leader Professor Ben Glocker, Daniel Coelho de Castro and Ian Walker. https://lnkd.in/edrqTKrc Causal reasoning can shed new light on the major challenges in machine learning for medical imaging: scarcity of high-quality annotated data and mismatch between the development dataset and the target environment. A causal perspective on these issues allows decisions about data collection, annotation, preprocessing, and learning strategies to be made and scrutinized more transparently, while providing a detailed categorisation of potential biases and mitigation techniques. Along with worked clinical examples, we highlight the importance of establishing the causal relationship between images and their annotations, and offer step-by-step recommendations for future studies.

i🚀 Excited to announce that our new paper, "Partial Annotation Learning for Biomedical Entity Recognition," is now published in the IEEE Journal of Biomedical and Health Informatics and available Open Access! 🎉 🔍 Named Entity Recognition (NER) plays a crucial role in supporting biomedical research. However, both manual and automatic data annotation processes often miss entity labels, degrading model performance. In our paper, we systematically explore partial annotation learning for Biomedical Named Entity Recognition (BioNER) and propose a novel model: TS-PubMedBERT-Partial-CRF. 📊 Our findings demonstrate that partial annotation methods significantly outperform full annotation models, especially under high missing entity rates. In fact, our model outperforms the PubMedBERT tagger by 38% in F1-score in these challenging scenarios. You can read the full paper in the "Early Access" section of IEEE Xplore: https://lnkd.in/dsW-q3E4 Kudos to my amazing collaborators Giovanni Colavizza Giovanni Colavizza and Zhixiong Zhang for their valuable contributions! 🙌 #BioNER #MachineLearning #AI #LLM

📃Scientific paper: Interpretable Medical Image Classification using Prototype Learning and Privileged Information Abstract: Interpretability is often an essential requirement in medical imaging. Advanced deep learning methods are required to address this need for explainability and high performance. In this work, we investigate whether additional information available during the training process can be used to create an understandable and powerful model. We propose an innovative solution called Proto-Caps that leverages the benefits of capsule networks, prototype learning and the use of privileged information. Evaluating the proposed solution on the LIDC-IDRI dataset shows that it combines increased interpretability with above state-of-the-art prediction performance. Compared to the explainable baseline model, our method achieves more than 6 % higher accuracy in predicting both malignancy (93.0 %) and mean characteristic features of lung nodules. Simultaneously, the model provides case-based reasoning with prototype representations that allow visual validation of radiologist-defined attributes. ;Comment: MICCAI 2023 Medical Image Computing and Computer Assisted Intervention Continued on ES/IODE ➡️ https://etcse.fr/W7s ------- If you find this interesting, feel free to follow, comment and share. We need your help to enhance our visibility, so that our platform continues to serve you.

1/ Our paper titled "Synthetic data in radiological imaging: Current state and future outlook" was recently accepted to the British Journal of Radiology (BJR), Artificial Intelligence. This is a joint work with Andreu Badal Jana Delfino Miguel Ángel Lago Ángel Brandon J. Nelson Niloufar Saharkhiz, PhD Berkman Sahiner Ghada Zamzmi aldo badano Link: https://lnkd.in/eFpJnta4 Synthetic data has many uses and great potential to enhance radiological AI applications. Synthetic data in radiology can be generated using statistical generative, physics-based modelling, or physics-informed models. Synthetic data can be used to mitigate various limitations of patient-based datasets, reducing associated risks, data acquisition time, cost, and complexity, and mitigating existing data imbalances. Overall, the choice of how to generate synthetic data depends on the availability and type of source of evidence used to create synthetic data, available biological or physics knowledge. Also, it is crucial to identify the patient data distribution the synthetic data is intended to represent.   #fda #regulatoryscience #osel #cdrh

I am thrilled to announce that our paper, "Generative Adversarial Networks in Medical Image Analysis: A Comprehensive Survey", has been officially published by Springer as part of the ICICC 2024 proceedings! 🚀✨ This paper explores the revolutionary role of Generative Adversarial Networks (GANs) in the field of medical imaging, highlighting how GANs are pushing the boundaries of diagnostic technologies and enhancing various applications such as: 🔹 Medical Image Segmentation – Accurately identifying and outlining structures like organs or tumors in medical images 🔹 Image Classification – Improving diagnostic precision by classifying medical conditions based on advanced imaging data 🔹 Image Reconstruction – Reconstructing clearer, high-quality images from incomplete or noisy medical data 🔹 Image Synthesis – Generating lifelike medical images that closely mimic real patient data, vital for training models and overcoming data scarcity 🔹 Noise Reduction – Enhancing the clarity and usability of medical images by reducing artifacts and improving quality 🔹 Anomaly Detection – Assisting physicians in identifying minute abnormalities in medical images for early disease detection Through these applications, GANs hold immense promise for improving disease detection, overcoming data limitations, and tailoring treatment approaches by generating highly authentic images that mirror real patient data while ensuring privacy protection. Moreover, the paper highlights the role of advanced deep learning techniques such as Conditional GANs (cGANs) for controlled image generation, CycleGANs for image-to-image translation, and Deep Convolutional GANs (DCGANs) to enhance the quality of generated images. These deep learning methods are essential in improving the accuracy and efficiency of medical image analysis and clinical decision-making. The paper also outlines future research directions, encouraging exploration of even more innovative uses of GANs in medical imaging to advance clinical practice and medical research to new heights. This achievement would not have been possible without the incredible support of my co-authors, Kancharagunta Kishan Babu and Sreeja Nukarapu. I'm proud of what we’ve accomplished together and excited to see where this research leads next! Check out the full paper here: https://lnkd.in/g29GXrFD #MedicalImaging #AI #GenerativeAdversarialNetworks #MedicalTechnology #HealthcareInnovation #ArtificialIntelligence #DataScience #Research #DeepLearning

🔬✨ CT-Less Organ Segmentation in PET: A Deep Learning Preprint ✨🔬 We’re excited to share a preprint led by Yazdan Salimi from Geneva University Hospital, under supervision from Prof. Habib Zaidi, and Dr. Ismini Mainta, proposing a groundbreaking approach to multi-organ segmentation in PET imaging without the need for CT. 🧠📊 🔹 Study Overview: Traditional segmentation in PET imaging relies heavily on CT, but CT-PET alignment issues and PET-only workflows can limit its effectiveness. This study introduces a CT-independent deep learning model using PET emission data alone, addressing these challenges head-on. Models and Data: Advanced nnU-Net models were trained and tested on a dataset of over 2,000 PET/CT images, focusing on 18F-FDG and 68Ga-PSMA tracers. Innovation: By eliminating CT reliance, this pipeline enables consistent, accurate organ segmentation, even in PET-only environments. 🚀 🔹 Key Findings: High Accuracy: Achieved Dice coefficients of 0.81–0.82 for 18F-FDG and 0.77–0.79 for 68Ga-PSMA, with strong performance in organs like the brain and lungs. Versatility: Supports applications in dosimetry, kinetic modeling, and radiomics, expanding the utility of PET imaging in clinical and research settings. GitHub Access: The authors have made their repository publicly available for exploration and feedback — try it out here: 🔗 https://lnkd.in/ejhkigVx 👏 We wish this preprint success in finding a peer-reviewed journal for publication and invite the community to explore and engage with this exciting new approach! 🔗 Read the preprint: https://lnkd.in/eC9HkqNg #NuclearMedicine #DeepLearning #PETImaging #MedicalImaging #ArtificialIntelligence #OrganSegmentation #SwissResearch #OpenScience

Hey everyone, I'm thrilled to announce the publication of my first research paper, "Advancing Medical Imaging Through Generative Adversarial Networks: A Comprehensive Review and Future Prospects," in the Cognitive Computation journal by Springer, which has an impressive impact factor of 4.3 (SCI-indexed). In collaboration with Bealu Girma, we delved into the transformative potential of Generative Adversarial Networks (GANs) in diagnostic imaging. Our work highlights how this cutting-edge technology can revolutionize the field of medical imaging. A special thanks to Associate Dr. Vikas Hassija, Prof. Vinay Chamola, and Aniruddha Mukherjee for their invaluable support and mentorship. 🙏 I'm deeply grateful to everyone who has supported me on this journey. This is just the beginning—more research papers are on the horizon as I continue to explore the exciting realms of Generative AI and ML. 🔗 Check out our paper: https://lnkd.in/gnYH5rqc Here's to a future brimming with innovation and positive impact! #ResearchPaper #MedicalImaging #GenerativeAdversarialNetworks #GANs #ArtificialIntelligence #AIResearch #HealthcareInnovation #CognitiveComputation #SpringerJournal #ScientificPublication #InnovationInHealthcare

Meeting online with Wendy J Erselius, from HARVARD MEDICAL SCHOOL, USA, for partnering on an international initiative to build public multi-institutional imaging datasets, named MAIDA (Medical ARTIFICIAL INTELLIGENCE Data for All). The overall goal of MAIDA is to bring together a diversity of patient populations from a large cohort of institutions on which to base the evaluation of AI algorithms. Ultimately, this will provide rigor to algorithms and open up access to multi-institution validation for research and innovation. I was invited after Harvard Medical School have recently come across our paper titled, " AI-driven deep CNN approach for multilabel pathology classification using chest X-Rays", published in PeerJ Computer Science. This initiative, curating datasets of chest X-rays, will enable the development of AI models that can be validated across diverse clinical settings. https://lnkd.in/dcahi_9i

Hey family, I am excited to share that my first research paper, titled "Advancing Medical Imaging Through Generative Adversarial Networks: A Comprehensive Review and Future Prospects," has been published in the Cognitive Computation journal by Springer, which boasts an impact factor of 4.3 (SCI-indexed). Collaborating with my colleague Abiy Mamo, we explored how Generative Adversarial Networks (GANs) can revolutionize diagnostic imaging. Our paper highlights the immense possibilities of this technology in reshaping medical imaging. Special thanks to Associate Dr. Vikas Hassija, Prof. Vinay Chamola, and Aniruddha Mukherjee for their support and mentorship. 🙏 Heartfelt gratitude to everyone who has been part of this journey. This is just the beginning—more research papers will be coming ahead as I continue to explore the exciting field of Generative AI and ML. 🔗 Check out our paper: https://lnkd.in/gnYH5rqc Feel free to inbox me for a free PDF. Here's to a future filled with innovation and positive impact! #ResearchPaper #MedicalImaging #GenerativeAdversarialNetworks #GANs #ArtificialIntelligence #AIResearch #HealthcareInnovation #CognitiveComputation #SpringerJournal #ScientificPublication #InnovationInHealthcare

🩻🔬New #aihealthcare research from Massachusetts Institute of Technology, Broad Institute of MIT and Harvard, and Massachusetts General Hospital are exploring #machinelearning frameworks to increase decision capabilities to help improve #imagesegmentation accuracy. Marianne Rakic, an MIT Computer Science PhD candidate serves as lead author on a paper illustrating the capabilities of a new medical imaging assistive AI tool named,Tyche, that composes multiple ambiguous image segmentations for the purpose of creating plausible label maps to help de-risk clinical analysis. “Ambiguity has been understudied. If your model completely misses a nodule that three experts say is there and two experts say is not, that is probably something you should pay attention to,” - Adrian V. Dalca Sr. Author, Assistant Professor Harvard Medical School & Massachusetts General Hospital, and research scientist with MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) #medicalinnovation #imagingscience #neuralnetworks The Mullings Group