Kimia Lab reposted this
It turns out cotton candy does not satisfy your hunger, gives you diabetes, and is ridiculously expensive. Great study by Nita Mulliqi et al. URL: https://lnkd.in/gcu6qkme #AI #LLMs #FoundationModels #Pathology
Kimia Lab
Research Services
Rochester, Minnesota 1,562 followers
Laboratory for Knowledge Inference in Medical Image Analysis
About us
The Laboratory for Knowledge Inference in Medical Image Analysis, short KIMIA Lab, established in January 2024 at Mayo Clinic, envisions to conduct research at the forefront of mass image data in medical archives using machine learning schemes with ultimate goal of extracting information that cannot only support a more speedy and accurate diagnosis and treatment of many diseases but also, and more significantly, establish new quality assurance based on mining of collective knowledge and wisdom. The acronym KIMIA happens to sound like the Greek word "χυμεία" (the art of alloying metals, or alchemy). Once upon a time, scientists assumed they can find techniques to turn any metal into gold. Perhaps the gold of the 21st century would be rather buried underneath (or among) the big image data.
- Industry
- Research Services
- Company size
- 2-10 employees
- Headquarters
- Rochester, Minnesota
- Type
- Educational
- Founded
- 2024
- Specialties
- Artificial Intelligence, DIgital Pathology, Computer Vision, Machine Learning, Image Search, Multimodal Retrieval, Computational Pathology , and Foundation Models
Locations
-
Primary
Get directions
200 1st St SW
Rochester, Minnesota 55905, US
Employees at Kimia Lab
Updates
-
Kimia Lab reposted this
In last month’s highlighted blog, Dr. H.R. Tizhoosh discusses the setbacks and challenges in AI model building and its lasting impacts on sensitive fields like healthcare and finance. As AI in digital pathology emerges as a key tool in practices, Dr. Tizhoosh points out the potential risks and costs associated with poorly constructed AI models and its potential to affect patients negatively. Read more: https://lnkd.in/gapxwYej
-
-
Kimia Lab reposted this
Just published: Validation of histopathology foundation models through whole slide image retrieval (23 organs, 117 cancer subtypes) URL: https://lnkd.in/gMQFZxJv Saghir Alfasly, Ghazal Alabtah, Sobhan Hemati, Krishna Rani Kalari, Joaquin Garcia, MD #AI #Pathology #Histopathology #DeepLearning #FoundationModels #Image #Search #Retrieval
-
-
Kimia Lab reposted this
Foundation models in pathology are not robust. Great investigation by Edwin D. de Jong et al. contributing to our understanding of what is wrong with foundation models URL: https://lnkd.in/eyqTenaH #Pathology #Foundation #Models #deeplearning #hhistopathology
-
-
Kimia Lab reposted this
-
WSI retrieval remains a challenging task...
Validating Foundation Models for Image Search in Histopathology URL: https://lnkd.in/gBE6JKJr #Pathology #Histopathology #Histology #DeepLearning #FoundationModel #UNI #Virchow #GigaPath #Search #Retrieval #Image #TCGA
-
-
Kimia Lab reposted this
Helping digital pathology scientists and businesses promote their products and services through (featured) blogs, podcasts and videos🔬Digital pathologist & publisher 👉🏻 Ask me about VIP sponsorship packages
Foundation pathology models brilliantly explained by H.R. Tizhoosh If you are tired of hearing about foundational models in pathology and not understanding what it's all about, I just found the perfect resource to end the confusion! I was doing some research for the next DigiPath Digest about foundation models and came across a lecture by H.R. Tizhoosh titled "Foundation Models and Information Retrieval in Digital Pathology " on YouTube. Key takeaways: 1) Foundation Models are trained on a lot of data - for pathology it would be over 1 mln whole slide images (better would be 5 mln... not patches, images) 2) I would need to be a zero-shot model (able to classify never seen data) 3) Retrieval augmented generation is more reliable for pathology purposes than generation alone. The best spent 15 min today Listen to the talk here: https://lnkd.in/eYC9SDQ3 and never be confused about those models EVER again! ------------------------------------------ #DigitalPathology #AIinHelathcare #FoundationModels
-
-
How to normalize sequences to be processed by LLMs... Great work by our Areej Alsaafin!
Just published: Harmonizing Immune Cell Sequences for Computational Analysis with Large Language Models URL: https://lnkd.in/gysYyHU6 Areej Alsaafin #AI #LLM #biological #sequencing #RNA #TCR #Tissue #pathology #omics
-
-
Select a few unique patches/tiles to represent a WSI ---------------------------------------------
Just published: SPLICE - Streamlining Digital Pathology Image Processing A new unsupervised algorithm for patching WSIs; - finally a better alternative to Yottixel's mosaic, the only existing patching method - Less redundant patches, - faster and more robust than Yottixel's mosaic - No labels needed; automatically select unique patches/tiles URL: https://lnkd.in/gZ6zSyYZ Code: https://lnkd.in/gKjf887k #WholeSlideImage #WSI #DigitalPathology #Patching #Tiling #Patch #Tile Areej Alsaafin, Peyman Nejat, Saghir Alfasly, Ghazal Alabtah, Jibran Ahmed Khan, Abubakr Shafique
-
-
A major validation process that took several months at Kimia Lab, incorporating a large team of collaborators, to provide valuable insights into image search in histopathology.
Just published: Analysis and Validation of Image Search Engines in Histopathology Over a period of six months, we performed the most comprehensive analysis and validation of search engines in histopathology. - Accuracy: we examined macro-average of F1-score for top-1 search results - Accuracy: we examined macro-average of F1-score for majority of top-3 search results - Accuracy: we examined macro-average of F1-score for majority of top-5 search results - Speed: We analyzed the theoretical upper bounds - Speed: We measured the actual search times - Storage: For the first time we examined and measured the total storage requirements (indexing overhead) - Storage: We introduced a new measure: indexing storage per WSI - Robustness: We measured failure times - And finally We ranked all relevant search engines. ## Results: - The accuracy of all search engines is still low - Some search engines are quite fast and efficient (e.g., Yottixel and BoVW) - Some search engines are very slow and inefficient (SMILY, RetCCL, SISH, HSDH,HSHR) - There are search engines that you can improve (e.g., Yottixel and BoVW can improve by using better embeddings, or SMILY could use a patching algorithm) - There are search engines that you cannot improve (e.g., SISH because of major design flaws, or RetCCL because it is a network that needs to be replaced) Download URL: https://lnkd.in/ga7z-A-Y #information #retrieval #search #image #pathology #histopathology #deeplearning #AI #foundationmodels Isaiah Lahr, Saghir Alfasly, Peyman Nejat, Jibran Ahmed Khan, Luke Kottom, Vaishnavi Kumbhar, Areej Alsaafin, Abubakr Shafique, Sobhan Hemati, Ghazal Alabtah, Nneka Comfere, Dennis Murphree, Aaron Mangold, Saba Yasir, Chady Meroueh, Lisa Boardman, Vijay Shah, Joaquin Garcia, MD
-