Clozel, T. | Search bioRxiv

bioRxiv2019

Transcriptomic learning for digital pathology

Deep learning methods for digital pathology analysis have proved an effective way to address multiple clinical questions, from diagnosis to prognosis and even to prediction of treatment outcomes. They have also recently been used to predict gene mutations from pathology images, but no comprehensive evaluation of their potential for extracting molecular features from histology slides has yet been performed. We propose a novel approach based on the integration of multiple data modes, and show that our deep learning model, HE2RNA, can be trained to systematically predict RNA-Seq profiles from whole-slide images alone, without the need for expert annotation. HE2RNA is interpretable by design, opening up new opportunities for virtual staining. In fact, it provides virtual spatialization of gene expression, as validated by double-staining on an independent dataset. Moreover, the transcriptomic representation learned by HE2RNA can be transferred to improve predictive performance for other tasks, particularly for small datasets. As an example of a task with direct clinical impact, we studied the prediction of microsatellite instability from hematoxylin & eosin stained images and our results show that better performance can be achieved in this setting.

bioinformatics↗

bioRxiv2018

Brain age prediction of healthy subjects on anatomic MRI with deep learning: going beyond with an \"explainable AI\" mindset

ObjectivesDefine a clinically usable preprocessing pipeline for MRI data\n\nPredict brain age using various machine learning and deep learning algorithms\n\nDefine Caveat against common machine learning traps\n\nData and MethodsWe used 1597 open-access T1 weighted MRI from 24 hospitals.\n\nPreprocessing consisted in applying : N4 bias field correction, registration to MNI152 space, white and grey stripe intensity normalization, skull stripping and brain tissue segmentation\n\nPrediction of brain age was done with growing complexity of data input (histograms, grey matter from segmented MRI, raw data) and models for training (linear models, non linear model such as gradient boosting over decision trees, and 2D and 3D convolutional neural networks).\n\nWork on interpretability consisted in (i) proceeding on basic data visualization like correlations maps between age and voxels value, and generating (ii) weights maps of simpler models, (iii) heatmaps from CNNs model with occlusion method.\n\nResultsProcessing time seemed feasible in a radiological workflow : 5 min for one 3D T1 MRI.\n\nWe found a significant correlation between age and gray matter volume with a correlation r = -0.74. Our best model obtained a mean absolute error of 3.60 years, with fine tuned convolution neural network (CNN) pretrained on ImageNet.\n\nWe carefully analyzed and interpreted the center effect.\n\nOur work on interpretability on simpler models permitted to observe heterogeneity of prediction depending on brain regions known for being involved in ageing (grey matter, ventricles). Occlusion method of CNN showed the importance of Insula and deep grey matter (thalami, caudate nuclei) in predictions.\n\nConclusionsPredicting the brain age using deep learning could be a standardized metric usable in daily neuroradiological reports. An explainable algorithm gives more confidence and acceptability for its use in practice. More clinical studies using this new quantitative biomarker in neurological diseases will show how to use it at its best.\n\nFOREWORDO_ST_ABSAbout OwkinC_ST_ABSOWKIN was co-founded in 2016 by Thomas Clozel, MD, a clinical research doctor and former assistant professor in clinical hematology and Gilles Wainrib, PhD, a pioneer in the field of Artificial Intelligence in biology. OWKIN passed the proof-of-concept phase and is now providing its innovative AI algorithms to several of the largest cancer centers and pharmaceutical companies in Europe and in the US. With offices in New York and Paris, we pride ourselves in building a company culture around transparency, collaboration, challenge, optimism and fun.\n\nOwkins teamOwkins team is international, multidisciplinary with incredible talent in machine learning, medicine and business. Our data scientists are among the best in the world, with several Kaggle Masters (top global 100), a DREAM Challenge top performer, and publications in ICML, NIPS and other top scientific journals.\n\nTasks repartitionIdea : Thomas Clozel, Roger Stup, Simon Jegou, Paul Herent\n\nBibliography : Paul Herent, Simon Jegou, Thomas Clozel\n\nData access : Simon Jegou, Paul Herent\n\nData cleaning : Simon Jegou, Paul Herent\n\nData analysis : Simon Jegou, Paul Herent\n\nData preprocessing : Simon Jegou, Paul Herent\n\nData analysis : Simon Jegou, Paul Herent\n\nTraining of models : Simon Jegou, Paul Herent\n\nWork on interpretability : Simon Jegou, Paul Herent\n\nWriting : Paul Herent\n\nRereading : Simon Jegou, Thomas Clozel, Julien Savatovsky, Roger Stupp, Olivier Elemento, Kim Gillier\n\nSubmission to medical congress : Paul Herent, Simon Jegou\n\nThanks to...\n\nSimon Jegou, for your mentoring in Machine Learning,\n\nThomas Clozel and Gilles Wainrib, for your welcome at Owkin, very benevolent, Roger Stupp, for your support and re-reading,\n\nJulien Savatovsky, for your support and re-reading,\n\nValentin Ame and Sylvain Toldo, for your help on the beautiful figures and the design of the related blogpost,\n\nAll the Owkin team members, for the great team work we did (and hope well do) between Paris and New York : Anna Huyghues Despointes, Anna I. Bondarenko, Pierre Courtiol, Derek T. Russell-Kraft, Cedric Whitney, Meriem Sefta, Vincent Lepage, Adrian Gonzalez, Maxime HE,Paul Jehanno, Raphael Leger, Alicia Simion, Eric Tramel, Mikhail Zaslavskiy, Pierre Manceron, Chloe Simpson, Paul Mabillot, Valentin Ame, Mathieu Galtier, Camille Marini, Sylvain Toldo, Charlie Saillard, Olivier Dehaene, Olivier Moindrot,\n\nPascal Roux, for your support, your help, your advices,\n\nAxelle, for your patience and support.

neuroscience↗