AI models trained on pathology slide images have the potential to predict diagnosis, prognosis, and new biomarkers. However, efforts to apply such models clinically have been hindered by a lack of clinical data from disease-specific cohorts and rare conditions. To address this gap, Tong Ding and Faisal Mahmood joined Harvard Medical School colleagues Sophia Wagner, Andrew Song, Richard Chen, and Long Phi Le to develop TITAN, a new model trained on more than 330,000 slide images and corresponding pathology reports. TITAN consistently outperforms other models at many clinical tasks, and highlights the utility of taking a vision-language approach to model training. Read more in Nature Medicine. 🔗: https://lnkd.in/gG5q-8XM #BroadInstitute #Science #ScienceNews #Research #ScientificResearch
Phenomenal! The future beckons.
Outstanding model
Amazing work!
Beautiful work! "We conjecture that some of these insights can be readily translated into other domains of pathology foundation models, such as hematopathology, spatial transcriptomics, 3D pathology and multiplex imaging." Bridging this gap seems of immense and immediate value. Clearly, we are amidst the early stages of a revolution in spatial biology that is enabling mapping of 1000s of RNAs (up to transcriptome scale) with single molecule resolution. On top of this, we at Proteintech Genomics are working on enhancing these capabilities to enable mapping of 100s-1000s of proteins in the same samples/slides. However promising, there is a sizeable elephant in the room. A major gap persists in how samples are handled, processed, and interpreted in research and clinical settings. If these models were (1) accessible to researchers and (2) capable of reliably identifying clinically relevant pathological cells/structures in high resolution multiomic spatial data from platforms like 10x Genomics' Xenium or Visium HD workflows (with or without H&E), this could fast track the high resolution mapping of the molecular underpinnings of the observed pathologies, thereby enhancing development of effective therapeutics. Onward!