Stumpe: Shows video. (Wow what production values Google $ can buy.) #AACR18
9:33am April 16th 2018 via Hootsuite
Stumpe: So bring AI to the microscope: from image to compute calc probabilities and output to display overlay. User can see AR image #AACR18
9:32am April 16th 2018 via Hootsuite
Stumpe: A new microscope: scanners are $, IT infrastructure, disruptive to existing workflows, not all clinical needs addressed #AACR18
9:31am April 16th 2018 via Hootsuite
Hipp: Onto Gleason grading Prostate Ca scoring 1-5, 3/4 and 4/5 are where subjectivity enters in (shows results from 8 pathologists) #AACR18
9:28am April 16th 2018 via Hootsuite
Hipp: Showed a time savings of 32%; and reduced error by 30%. #AACR18
9:26am April 16th 2018 via Hootsuite
Hipp: Asked pathologists to highlight, and cp to AI result. Feasibilty study: 6 pathologists w/70 samples, 24 neg, 14 micromet, 14+ #AACR18
Stumpe: tile image into ~1M tiles; binary classify, use Inception v3. Similar to image recognition for identifying dogs in images #AACR18
9:23am April 16th 2018 via Hootsuite
Stumpe: 15Gb pixels - ~1000 DSLR photos. 100k x 100k pixels. Google has AI expertise; TensorFlow, mapping, cars. #AACR18
9:22am April 16th 2018 via Hootsuite
Stumpe: Detection is hard due to germinal centers, macrophages, that mimic tumor appearance. Shows similarity between them #AACR18
9:21am April 16th 2018 via Hootsuite
Stumpe: Looking at metastases in lymph nodes in breast cancer - a needle in a haystack. Image is 150K x 90k pixels. Tumor 1300x300 #AACR18
9:20am April 16th 2018 via Hootsuite
Stumpe: Detecting metastatic br ca in lymph nodes w/AI; AI-based PrCa Gleason grading; an a new instrument. #AACR18
9:19am April 16th 2018 via Hootsuite
Jason Hipp, Martin Stumpe (Google CA) Advancing cancer diagnostics with artificial intelligence #AACR18
9:18am April 16th 2018 via Hootsuite
Barker: #AACR18 What we need from oncology: we need 'gold-std databases' AI cannot work with poor-quality data.
8:54am April 16th 2018 via Hootsuite
Barker: #AACR18 But you need to have the right question. Also information theory: the nature of information. Information is not data.
8:52am April 16th 2018 via Hootsuite
Barker: #AACR18 Big data - volume, velocity, volume, and major data types are increasing. The future is deep learning, machine learning, AI
Barker: Occur in 3D space over time; biomarkers must reflect complex contexts at different scales over time #AACR18
8:47am April 16th 2018 via Hootsuite
Barker: #AACR18 Cancer a complex system, produce emergent properties like metastasis. Chemical, virus, hormone, nutrition: mutations accum.
8:44am April 16th 2018 via Hootsuite
Barker: ImmunoRx - shows Wagle slide - resistance evolution. Complex adaptive systems; robust, evolvable, adaptable all interacting #AACR18
8:43am April 16th 2018 via Hootsuite
Barker: #AACR18 The future: if each pt is unique, 'malignant snowflakes': each cancer carries multiple unique mutations
8:42am April 16th 2018 via Hootsuite
Barker: #AACR18 Stds / evidence 'needed at each step'. Three pathways through regulatory approval: scientific comm consensus; drug-spec dev
8:41am April 16th 2018 via Hootsuite
Barker: 'Lack of data stds is killing us' Pipe: early discoery, translatable disc, assay dev, assay perf, biomarker qual, validation #AACR18
Barker: Inadequate exp design; poor quality specimens; lack of tech stds; poor quality data/lack of data stds; inadquate analytics #AACR18
8:39am April 16th 2018 via Hootsuite
Barker: #AACR18 The org: https://t.co/ajfsJl8QJq mission: for biomarker discovery. 6 'showstopping barriers' - lack of clin relevent Q's
Barker: Ameneble to dev into simple-to-use tech; cost-effective; must fit easily into flow to clin care. NBDA non-profit setup #AACR18
8:37am April 16th 2018 via Hootsuite
Barker: Why biomarkers fail: integrative to address complexity; adaptive to track dynamic nature; ubiquitous; simple; sample-frienly #AACR18
Barker: If we had high value, biologically and clin relevant biomarkers - smarter study design, less time, fewer volunteers needed #AACR18
8:36am April 16th 2018 via Hootsuite
Barker: Oncology clinical trials: massive attrition, long duration, high costs. 10,000:1 chance of success; 12-15y, $2B #AACR18
8:35am April 16th 2018 via Hootsuite
Barker: 150K claimed biomarkers; est 100 biomarkers routinely used. ref '11 https://t.co/KR8V8kOJ4h CDx - not many. Proteins: 1/yr #AACR18
8:34am April 16th 2018 via Hootsuite
Barker: Now allowed subtypes, major pathways, and a starting place to understand how pathways interact. But treatment hasn't changed #AACR18
8:33am April 16th 2018 via Hootsuite
Barker: Shows image from first GBM TCGA paper - the kind of science that changes the world. GBM - complex and deadly. #AACR18
8:32am April 16th 2018 via Hootsuite
Barker: Patient volumes can approach 1000 PB per year at maturity. $500. Or lower now. 'The data is the best we've ever had.' #AACR18
8:31am April 16th 2018 via Hootsuite
Barker: Now with the 'Omics revolution': she started TCGA, after HGP now dominating all of our thinking. In the millions of WGS #AACR18
8:30am April 16th 2018 via Hootsuite
Barker: BP correlates well with CV health/disease status; easy and inexpensive to assess. Low vs high risk of CV event #AACR18
8:29am April 16th 2018 via Hootsuite
Barker: Blood pressure is an ideal biomarker: it measures the integrated expression of a complex, adaptive system. Quantitative info #AACR18
8:28am April 16th 2018 via Hootsuite
Barker: Woodcock quote: 'Why can't you tell me if my medicine is working? ...Biomarkers are the key tools that we need' 2014 NBDA #AACR18
8:27am April 16th 2018 via Hootsuite
Barker: Clinically useful biomarkers are almost always complex; biology is complex. Biomarkers have 'incredible potential'. #AACR18
8:26am April 16th 2018 via Hootsuite
Barker: The FDA doesn't care if it's a 'black box' - as long as it is consistent, prognostic etc. #AACR18
8:25am April 16th 2018 via Hootsuite
Barker: Imaging: the best biomarkers we have right now (MRI, CT, FDG-PET). Complex biomarkers are informed algorithms. The Future #AACR18
Barker: Precision medicine has moved along b/c NGS has been getting better; proteomics 'has not done well at all' #AACR18
8:24am April 16th 2018 via Hootsuite
Barker: Safety con't: b/c we lack toxicity biomarkers. Some biomarker 'classes' - Genomics produces biomarkers, GEx, proteomics... #AACR18
Barker: Pharmacodynamic - probably the most impt, the biol response following Rx intervention. Safety - we often overtreat... #AACR18
8:23am April 16th 2018 via Hootsuite
Barker: Cancer biomarkers can be classified according to their clinical utilty; disease risk, Dx, monitoring, prognostic, predictive #AACR18
8:22am April 16th 2018 via Hootsuite
Barker: Precision - 'we've lost, in terms of biomarkers...' Fit for Purpose: FDA term, defines the context of use. ROC - familiar #AACR18
Barker: The value of biomarkers often lie in ability to meet criteria 'defined by the following...' (long list - Sens, Spec...) #AACR18
8:20am April 16th 2018 via Hootsuite
Barker: #AACR18 (I believe she quoted from this BEST document via FDA: https://t.co/T95C3I3Gqe ) Prentice criteria from '89
Barker: With immunoRx, checkpoint inh are only one piece of a complex system... '15 FDA redefined definition of a biomarker #AACR18
8:17am April 16th 2018 via Hootsuite
Barker: 10y at the NCI, headaches of biomarkers; 'today, the truth is we have almost none'. Alt title 'looking in the wrong places' #AACR18
8:16am April 16th 2018 via Hootsuite
Anna Barker (Arizona State Univ) Cancer Biomarkers - Moving from Promise to Reality #AACR18
8:14am April 16th 2018 via Hootsuite
@DCcloudy Agreed!
8:09am April 16th 2018 via Hootsuite in reply to DCcloudy
Ladas: Detecting MSI via PCR generate 'stutter peaks'. So the DSN (duplex-spec nuclease) method digests dsDNA pref over ssDNA, muts #AACR18
5:39pm April 15th 2018 via Hootsuite
