440 likes | 668 Views
Flagship Biosciences LLC. Validation of digital pathology applications in regulated study environments. Digital Pathology in the News. CAP 2010 ‘Digital pathology continues to generate industry buzz….’
E N D
Flagship Biosciences LLC Validation of digital pathology applications in regulated study environments
Digital Pathology in the News CAP 2010 ‘Digital pathology continues to generate industry buzz….’ ‘there are over a dozen FDA 510(k) clearances for digital analysis of immunohistochemistry procedures, the waiting game continues for how the agency wants to regulate digitalization of hematoxylin and eosin (H&E) slides using whole slide imaging (WSI) systems’ ‘once these regulatory barriers are negotiated, digital pathology will move ahead at breakneck speed’
New Technologies for Health Care • Star Trek technologies • VISOR • Hypospray • Tricorder • The holy grail of medicine • Digital Radiology • Digital Pathology Are new technologies outpacing regulatory guidance? Who are the guiding decision-makers?
Regulatory Needs in Digital Pathology? • Use of whole slide images in an electronic environment – from acquisition to storage • Systems qualifications (IQ/OQ/PQ validation) • Quantitative image analysis on whole slide and TMA images • Accessioning, viewing, scoring by pathologists, and adjudication • Peer reviews and digital archiving
Regulatory & Compliance Digital Pathology in Drug Development Clinical Novel regulatory problems? Preclinical Discovery
Regulatory Guidance • Regulatory requirements for digital pathology present a complex series of processes in the drug development process • Digital images • Storage • Annotations • Image analysis • www.hhs.gov or www.fda.gov • CFR - Code of Federal Regulations Title 21 (Food and Drugs) • PART 11 Electronic Records; Electronic Signatures • PART 58 Good Laboratory Practice for Nonclinical laboratory Studies • 501(K) Premarket Notification • In Vitro Diagnostic Multivariate Index Assays (21 CFR 809.3) • CLIA - Clinical Laboratory Improvement Amendments
Digital Pathology and IA Discovery • IHC investigations in potential new target organs • Researchers seeking to validate hypothesis • Verification and replication of literature claims • Tissues from commercial tissue banks have unknown demographics, outcomes, unknown pre-analytical variables, etc • Xenograft modeling • In vivo pathobiology studies • Early efficacy studies
Digital Pathology and IA Preclinical • Toxicology studies • Safety • Efficacy • Pharmacokinetic • Special studies • Peer review • Veterinary toxicological pathologists • North America, Japan, Europe (England, Germany, France, Switzerland) • Few overseas - especially in emerging biotech areas such as India and China • VIPER
Digital Pathology and IA Clinical • Clinical trials • Inclusion criteria • Retrospective analysis • Companion DX • Selection of biomarkers • Kit development • Pathology scoring • Treatment regimens for personalized medicine • HER2, ER, PR – breast cancer • EGFR – lung cancer (NSCLC) • Multiplexing multiple biomarkers (IHC-based)
Multiplexing Multiple Markers on One Slide is Difficult Quantum Dots …ready for the clinic…next year • Tough problem Dual-stained IHC slides • Great research tool, double-staining is generally not high quality enough to run in diagnostic settings • Problems with cross-reactivity between chromogens, avoid DAB • US Labs TriView for prostate and breast – for color aid for pathologist, not quantitation • Breast: CK 5/6 (cytoplasmic brown) and p63 (nuclear/brown) stain myoepithelial cells, while CK8/18 labels the cytoplasm (cytoplasmic/red) of ductal or lobular epithelium. Dual or triple stained immunofluorescent (IF) slides • Expensive, no anatomical tissue context • IF not used extensively in the clinic
Multiplexing Biomarkers in Tissue Sections Single section Multiple sections Slide not preserved Slide preserved FACTS Flagship Industry Layered IHC 20/20 GeneSystems Brightfield AQUA HistoRx IHC slide IHC slide Fluorescence IHC slide IHC slide Sequential Imaging GE IHC slide Q Dots Ventana IHC slide IHC slides
FDA Protein Expression Clearances Date510(k) Number Tissue Stain Reagent Application ScanScope XT System (Aperio) 2009/08 K080564 Breast Her2/neu Dako Tunable Image Analysis - System 2008/10 K080254 Breast PR Dako Reading on Monitor - System 2008/08 K073667 Breast ER/PR Dako Image Analysis - System 2007/12 K071671 Breast Her2/neu Dako Reading on Monitor - System 2007/10 K071128 Breast Her2/neu Dako Image Analysis - System PATHIAM (Bioimagene) 2009/02 K080910 Breast Her2/neu Dako Image Analysis - System 2007/02 K062756 Breast Her2/neu Dako Image Analysis - SW VIAS (Tripath) 2006/09 K062428 Breast P53 Ventana Image Analysis - System 2006/04 K053520 Breast Ki-67 Ventana Image Analysis - System 2005/08 K051282 Breast Her2/neu Ventana Image Analysis - System 2005/05 K050012 Breast ER/PR Ventana Image Analysis - System ARIOL (Applied Imaging) 2004/03 K033200 Breast ER/PR Dako Image Analysis - System 2004/01 K031715 Breast Her2/neu Dako Image Analysis - System ACIS (Clarient/Chroma Vision) 2004/02 K012138 Breast ER/PR Dako Image Analysis - System 2003/12 K032113 Breast Her2/neu Dako Image Analysis - System QCA (Cell Analysis) 2003/12 K031363 Breast ER Dako Image Analysis - SW www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPMN/pmn.cfm
Multiple EMT IHC Biomarkers H&E E-cad Vim
FACTS*Feature Analysis on Consecutive Tissue Sections A multiplexing biomarker approach for analysis *Patent Pending
Automating quantitative IHC ROI analysis in tissue is a HARD problem… • What works on a few samples doesn’t translate to real-world samples, especially in clinical trials where the ability to control sample acquisition, handling, fixation, IHC, and scanning is limited • IHC histologies simply do not have enough biology information to allow the computer to quickly build a reproducible, reliable system • Tissue variability is difficult • on any computer software Where is my ROI?
Common IA Needs Neurofibrillary tangles & tau Neurology Amyloid plaque Liver toxicologies Spleen red/white pulp Biomarkers in kidney glomeruli Toxicology Beta cell mass in islets with stereology Beta cell mass in islets Diabetes Xenograft tumor / normal / necrosis Clinical trials samples tumor / normal / necrosis Tumor bank samples tumor / normal / necrosis Oncology Easy Impossible Robust Difficult
Feature Analysis on Consecutive Tissue Sections (FACTS) • Consecutive • tissue • sectioning 2. Automated feature recognition 3. Image and ROI registration 4. QC and pathologist review
Consecutive • tissue • sectioning • GOAL: Minimal disruption to histology lab processes • Careful sectioning to get excellent consecutive tissue ribbons • Control pre-analytical factors *All slides for biomarkers must be taken in same session
Feature Analysis on Consecutive Tissue Sections (FACTS) • Consecutive • tissue • sectioning Biomarker -1 Biomarker -2 1a. Slide staining H&E Biomarker -3 Biomarker -4
Consecutive • tissue • sectioning 2. Automated feature recognition GOAL: Optimal reproducible and scalable whole slide feature analysis • Automatically recognizing features with assist of special stains • Special stain examples: • Oncology: Tumor / stroma / necrosis differentiation • Prostate & Lung substructures • Diabetes / Pancreas: anti-insulin antibody for islets • Kidney / renal tox: glomeruli stains
Consecutive • tissue • sectioning 2. Automated feature recognition 3. Image and ROI registration GOAL: Successfully register image with <3% error rate on ROI transfers between consecutive sections • Image registration approaches from radiology • Multi-modal, semi-automatic approach • Requires first rotating, translating, and sizing two whole slide images • Secondary step involves transferred ROI alignment (rotating, translating, sizing approach to near boundaries)
Consecutive • tissue • sectioning 2. Automated feature recognition 3. Image and ROI registration 4. QC and pathologist review GOAL: Increase analysis accuracy while improving pathologist productivity • Technician review and exclusion of poorly identified features • Features missing in adjacent sections (e.g. end-cut glomeruli or islets) • Non-specific staining impacting feature recognition • Poorly matched features • Pathologist review and sign-out
Validation Approach • H&E stained slides were cut in 4 um sections. One section was used as the reference section. FACTS was run across consecutive sections and error analyses were calculated False negative area False positive area To estimate error per feature (as in this glomeruli example), we first must map the transferred region as well as find the “correct” region. The “correct” region can either be drawn manually, or using automated feature recognition, depending on the application. The differences between the two regions (XORed area) is then divided by the mapped region to give the percent error per feature
Kidney - Glomeruli Total error = 5.8% False positive = 0.9% False negative = 4.9% Ave diameter = 61 um 3.4% error 3.0% error 11.8% error 6.2% error 3.7% error 8.0% error
Pancreas - Islets 5.7% error 6.6% error 8.1% error 3.8% error Total error = 4.8% False negative = 3.8% False positive = 1.0% Ave diameter = 135 um
Liver – Bile Ducts 46 um 4.0% error 4.8% error 9.7% error 14.4% error Total error = 4.7% False negative = 4.1% False positive = 0.6% Average diameter = 56 um 13.1% error
Spleen – Peri-arteriolar Lymphoid Tissue 500 um 0.4% error 0.9% error 0.6% error Total error = 1.0% False negative = 0.7% False positive = 0.3% Average feature diameter = 520 um 1.6% error
Xenografts – Specific Area Selection Total error = 1.0% False negative = 0.4% False positive = 0.6% Average feature diameter = 490 um
Oncology Clinical Trials - NSCLC • Consecutive sections from NSCLC patients were cut and stained for an epithelial marker as well as a biomarker of interest. • Automated feature recognition run on the epithelial stain Normal bronchioles excluded manually Epithelial stain delineates tumor Staining of epithelial surface linings and normal alveolar tissue excluded programmatically
Oncology Clinical Trials - NCSLC • Automated feature extraction followed by vectorization to generate regions of interest - eliminates ‘non-alike’ tissue regions
Oncology Clinical Trials - NCSLC • Image alignment followed by ROI alignment • ROI transfer with human annotated areas for error calculations ROI alignment Image alignment on deconvolved hemotoxylin channels Total error = 3.2% False negative = 1.7% False positive = 1.5% Average feature diameter = 315 um
What is the limit on multiplexing? • 9 consecutive 4 µm sections from xenograft tumor • H&E staining • FACTS false positive and false negative rates
Advantages of FACTS • Multiple IHC biomarkers can be developed into one IVDMIA • More reliable approach for highly variable samples seen in real world situations • Cost-effective and fits well into current GLP and CLIA practice • No novel double/triple stains or biomarker development required • Full audit trail of glass slides • Follows a precedent path with standard brightfield IHC IA digital imaging 510k approval process
Regulatory Alignment of FACTS • Trackable, reproducible image transfer and registration • Similar process as precedent FDA clearances • Requires no novel histology processes • Review and pathologist sign out is the same • Validation through FDA regulations and CLIA compliance
Preclinical Toxicology Liver – bile ducts Kidney: glomeruli dysfunction Pancreas: islets, alpha/beta cell mass Spleen: red / white pulp, EMH Discovery & Clinical Multiple IHC measurements in xenografts IVDMIA development in lung samples Stroma / Cancer in ER/PR/HER2 TMA multiplexing in discovery and retrospective clinical trials PrognosDx epigenetic markers (5 histone markers) Ongoing Flagship Projects with FACTS
Steve Potts Trevor Johnson David Young Scott Watson Frank Voelker Erik Hagendorn Rob Diller Rob Keller Contact us at: pathservices@flagshipbio.com Dave@flagshipbio.com www.flagshipbio.com