Day 1 - Thursday - 0830 AM - Rick Avila

Documents

prevent-cancer-foundation
Controlling Variability in Lung Cancer Response Assessment Ricardo S. Avila May 13, 2010 Therapy Assessment Characteristics •Late stage •Thick Slice CT Assessment •Tumor response •ID new lesions ? Tumor Size 4 cm lesion ∆ t ? Time Start Therapy Assess Response RECIST 8mm ∆ D, 13 pixels 73% ∆ Volume Target Lesion Measurement RECIST: Sum of LD Progressive Disease ∆ D = +20% Unaided Interpretation Stable Disease weeks 4cm lesion ∆ D = -30% Partial Response Time Baseline & Treat Complete Response Erasmus et. al., JCO 2003 Intra-observer error PD: 9.5% of tumors PR:3% of tumors Inter-observer error PD: 30% of tumors PR: 14% of tumors Assess Response We Can Do Better Target Lesion Measurement RECIST: Sum of LD Progressive Disease Improve •Accurac y •Precisio n To Improve •Interval (Dt) •Study N Aided 3D Interpretation • ∆ t Time Early Detection & Nodule Sizing Complete Response Partial Response Stable Disease 4cm lesion Detecting a 50 Micron Displacement Patton and Byron Nature Reviews Drug Discovery 2007 Computed Tomography Siemens Emotion 16 16 Slice Scanner 1.00mm Slice Thickness B30s Kernel B60s Kernel GE LightSpeed Ultra 8 Slice Scanner 1.25mm Slice Thickness Very Low Dose Low Dose Measurement Challenges  Patient/Lesion Presentation – Size – Complexity – Changes over time (necrosis) –  Scanners – Hardware (collimation) – Software (releases) –  Protocols – ScanRx – Contrast – Patient position –  Observer – Seed points/ROI – Data Interpretation 5mm 2.5mm Volumetric Algorithm Challenges Boundary Identification Challenges No/Small ∆ I •Vascular network (Ev) •Bronchial network (Eb) •Pleura (Ep) •Sub-voxel edge (Es) Errors at 2 time points Ev Volumetric error strongly depends on lesion size and slice thickness Ep Es Pl e ur a Technical Focus Areas • Open Image Archives – LCA’s Give-A-Scan Project – OSA’s Interactive Science Publishing – RSNA’s Ad Hoc Committee on Open Image Archives  • Understanding Measurement Performance – – – – – Benchmarks: NIST Biochange and Volcano QIBA: Phantom Data Studies QIBA: Measurement Performance on Clinical Data Kitware Pocket Phantom • Open Source Algorithms and Models – Lesion Sizing Toolkit – COPD Modeling and Quantification – • Establishing Standards for Clinical Trials – QIBA: Volumetric CT Profiles Quantitative Identification of Patient Sub-Populations  • Analysis of imaging and clinical data can potentially identify patient populations that respond more favorably to lung cancer therapy  – Drug Efficacy • Lung Damage Assessment – COPD impacts aerosolized drug delivery – Lung Cancer Risk • – Safety • Cardiovascular damage Lung Cancer Alliance’s Give-A-Scan Project A Lung Cancer Alliance Project • Pilot project started in 2008 • Process and procedures were created for accepting and anonymizing datasets • ~30 individuals expressed interest in participating • 17 scans received, but 2 were not readable • Over 6 GB of image and meta data was collected • 9 patient scans have been prepared for public dissemination on a LCA website. Give-A-Scan Website Dataset includes: Age Gender Cancer Type Cancer Stage Family History 4 of the 9 subjects are never smo Legal Documents A large amount of effort spent on developing the legal framework • Informed Consent • End User License   An open set of legal resources for open image archives would benefit many projects  New CT Pocket Phantom New CT Pocket Phantom   Goal: To characterize the fundamental imaging characteristics of CT acquisitions performed in the Roche ABIGAIL study – 3D Resolution & Sampling Rate – Noise Characteristics – X-ray Attenuation Performance  Acrylic Delrin Teflon Urethane New CT Pocket Phantom Manufactured 21 phantoms and deployed them into the Abigail phase II clinical trial   Fully Automated Phantom Analysis Several Studies Underway Resolution vs. Distance to Isocenter In-Plane PSF σ σ = 0.53 mm D = 112 mm σ = 0.45 mm σ = 0.47 mmσ = 0.54 mm D = 49 mm D = 62 mm D = 118 mm σ = 0.53 mm = 0.45 mm σ = 0.44 mm σ = 0.51 mm σ D = 114 mm D = 43 mm D = 32 mm D = 104 mm Standard Kernel Bone Kernel Lung Kernel Comparison of the New Pocket Phantom with a Catphan Phantom  Calibration Study – Siemens Sensation 64 CT Scanner – 6 pocket phantoms placed in/near an anthropomorphic chest phantom – Catphan phantom also scanned – Varied slice thickness, mA, kVp, and pitch –  Pearson’s Correlation Coefficients – CT Density = 0.999 (P < 0.001) – Noise = 0.940 (P < 0.001) – Resolution = 0.929 (P < 0.001)  Open Source Lesion Sizing Toolkit The Lesion Sizing Toolkit http://public.kitware.com/LesionSizingKit/ The Lesion Sizing Toolkit (LST) is a free and open source software architecture designed to accelerate the development and evaluation of quantitative lesion sizing algorithms. Developed in 2008 Focused on Dissemination in 2009 RSNA Quantitative Reading Room of the Future Showcase Open Source Medical Imaging Software Course Benchmarks Volcano 2009 OSA ISP Special Issue on Imaging for Early Lung Cancer Detection Lung Cancer Risk Lung Cancer Formation • Significant tissue damage occurs as a result of particulate matter (PM) deposition  Hyaline Cartilage • Deposition is a function of air flow dynamics and PM characteristics • • Histology and CFD has shown up to a 100x greater PM deposition at: – Airway bifurcations  [Broday, Aerosol Science and Tech. 2004] – Respiratory bronchioles  [Churg & Brauer, Ultrastructural Path. 2000]  • Bifurcation and peripheral lung tissues likely exhibit some of the earliest preneoplastic changes in response to PM exposure Balashazy et al., J Appl Physiol 2003. Bifurcation Calcification in HRCT Bifurcation Calcification Open Image Archive 1.25mm Slice Thickness w/ Bone K Lung Cancer Risk Index (LCRI)   Features 1. Bifurcation Damage Index (BDI) • HRCT w/ B60f edge enhancing kernel • Mean of 5 airway bifurcations (~20min) • 2. FEV1/FVC • Decline associated with lung cancer risk • Follow ATS spirometry guidelines  BD CD  Classifier Method is Independent of Age, Gender, Pack Years… • Linear BDI vs. FEV1/FVC Hi gh er Hi gh er Regression line is for cancer cases scanned at 1mm slice thickness and FEV1/FVC > 55% Lo w er Lo we r Initial Performance Analysis Dataset Conditional Logistic Regression Cochran-Mantel-Haenszel (Odds ratio for a ∆ 0.033 in LCRI with (crude estimate) 1:3 matching ) 108 Cases Full Dataset OR = 1.84 CI: 1.18-2.85 p-value = 0.0067 OR = 2.89 CI: 1.02-8.19 p-value = 0.0467 67% sensitive 72% specific 79 Cases 1mm Only 100% sensitive 74% specific Conclusion: Individuals with higher LCRI are more likely to have lung cancer Data on 21 Cancers and 121 Controls COPD Pr ox im al Di st al Lu ng COPD Ca nc er Lung Cancer Risk Findings • Investigating a new quantitative imaging biomarker • Airway bifurcations are calcifying in a relationship with FEV1/FVC • In control cases, a significant trend observed between LCRI and age*pack years (P = 0.006) • Odds Ratio for LCRI is better than FEV1/FVC – LCRI = 2.73 (CI: 1.35-5.51, P = 0.005) = 0.44 (CI: 0.24-0.83, P = 0.005) – FEV1/FVC • Opportunities exist to identify new lung cancer patient sub-populations  Give-A-Scan Patient Donated Dataset Never Smoker, Cancer at 62, FEV1/FVC=84% Right 2.5mm Scan Standard Kernel Left Measuring Progress Interim Meetings 7 Workshops since 2004 1 Interim COPD Meeting Annual Workshop PCF/Cornell Database NCIA Give-A-Scan COPDGene? Large Open Image Databases QIBA FDA NIST Standards & FDA Approval Reproducibil ity & Comparison Early Clinical Trials BioChange & Volcano QIBA Studies Accelerate Developme nt of Therapy Assessmen t Methods Algorithms & Reference Methods Open Source Lesion Sizing Toolkit CT COPD Algorithms Publications Oncology Workshop Reports Quantitative CT Monograph ISP Oncology Special Issue Thank You Lung Cancer Risk Index Cancers and Age & PY Matched Controls (+/-10) PY = Age = 5 64 15 51 20 59 28 51 30 57 40 48 45 58 59 60 60 69 63 54 66 64 68 57 72 68 74 75 62 92 Cancer Subjects Sorted by Increasing Pack Years A control case was permitted to be used for more than 1 cancer case Lung Cancer Risk Index Cancers and Age & PY Matched Controls 1.0 mm CT Thickness Threshold 1.25 mm CT Thickness Threshold PY = Age = 5 64 15 51 20 59 28 51 30 57 40 48 45 58 59 60 60 69 63 54 66 64 68 57 72 68 74 75 62 92 Cancer Subjects Sorted by Increasing Pack Years A control case was permitted to be used for more than 1 cancer case (We are now using FEV1/FVC before bronchodilator) New Study Results (We are now using FEV1/FVC before bronchodilator) Thymom a New Study Results Carcinoid of the Thymus AAH
Please download to view
1
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Description
Text
Controlling Variability in Lung Cancer Response Assessment Ricardo S. Avila May 13, 2010 Therapy Assessment Characteristics •Late stage •Thick Slice CT Assessment •Tumor response •ID new lesions ? Tumor Size 4 cm lesion ∆ t ? Time Start Therapy Assess Response RECIST 8mm ∆ D, 13 pixels 73% ∆ Volume Target Lesion Measurement RECIST: Sum of LD Progressive Disease ∆ D = +20% Unaided Interpretation Stable Disease weeks 4cm lesion ∆ D = -30% Partial Response Time Baseline & Treat Complete Response Erasmus et. al., JCO 2003 Intra-observer error PD: 9.5% of tumors PR:3% of tumors Inter-observer error PD: 30% of tumors PR: 14% of tumors Assess Response We Can Do Better Target Lesion Measurement RECIST: Sum of LD Progressive Disease Improve •Accurac y •Precisio n To Improve •Interval (Dt) •Study N Aided 3D Interpretation • ∆ t Time Early Detection & Nodule Sizing Complete Response Partial Response Stable Disease 4cm lesion Detecting a 50 Micron Displacement Patton and Byron Nature Reviews Drug Discovery 2007 Computed Tomography Siemens Emotion 16 16 Slice Scanner 1.00mm Slice Thickness B30s Kernel B60s Kernel GE LightSpeed Ultra 8 Slice Scanner 1.25mm Slice Thickness Very Low Dose Low Dose Measurement Challenges  Patient/Lesion Presentation – Size – Complexity – Changes over time (necrosis) –  Scanners – Hardware (collimation) – Software (releases) –  Protocols – ScanRx – Contrast – Patient position –  Observer – Seed points/ROI – Data Interpretation 5mm 2.5mm Volumetric Algorithm Challenges Boundary Identification Challenges No/Small ∆ I •Vascular network (Ev) •Bronchial network (Eb) •Pleura (Ep) •Sub-voxel edge (Es) Errors at 2 time points Ev Volumetric error strongly depends on lesion size and slice thickness Ep Es Pl e ur a Technical Focus Areas • Open Image Archives – LCA’s Give-A-Scan Project – OSA’s Interactive Science Publishing – RSNA’s Ad Hoc Committee on Open Image Archives  • Understanding Measurement Performance – – – – – Benchmarks: NIST Biochange and Volcano QIBA: Phantom Data Studies QIBA: Measurement Performance on Clinical Data Kitware Pocket Phantom • Open Source Algorithms and Models – Lesion Sizing Toolkit – COPD Modeling and Quantification – • Establishing Standards for Clinical Trials – QIBA: Volumetric CT Profiles Quantitative Identification of Patient Sub-Populations  • Analysis of imaging and clinical data can potentially identify patient populations that respond more favorably to lung cancer therapy  – Drug Efficacy • Lung Damage Assessment – COPD impacts aerosolized drug delivery – Lung Cancer Risk • – Safety • Cardiovascular damage Lung Cancer Alliance’s Give-A-Scan Project A Lung Cancer Alliance Project • Pilot project started in 2008 • Process and procedures were created for accepting and anonymizing datasets • ~30 individuals expressed interest in participating • 17 scans received, but 2 were not readable • Over 6 GB of image and meta data was collected • 9 patient scans have been prepared for public dissemination on a LCA website. Give-A-Scan Website Dataset includes: Age Gender Cancer Type Cancer Stage Family History 4 of the 9 subjects are never smo Legal Documents A large amount of effort spent on developing the legal framework • Informed Consent • End User License   An open set of legal resources for open image archives would benefit many projects  New CT Pocket Phantom New CT Pocket Phantom   Goal: To characterize the fundamental imaging characteristics of CT acquisitions performed in the Roche ABIGAIL study – 3D Resolution & Sampling Rate – Noise Characteristics – X-ray Attenuation Performance  Acrylic Delrin Teflon Urethane New CT Pocket Phantom Manufactured 21 phantoms and deployed them into the Abigail phase II clinical trial   Fully Automated Phantom Analysis Several Studies Underway Resolution vs. Distance to Isocenter In-Plane PSF σ σ = 0.53 mm D = 112 mm σ = 0.45 mm σ = 0.47 mmσ = 0.54 mm D = 49 mm D = 62 mm D = 118 mm σ = 0.53 mm = 0.45 mm σ = 0.44 mm σ = 0.51 mm σ D = 114 mm D = 43 mm D = 32 mm D = 104 mm Standard Kernel Bone Kernel Lung Kernel Comparison of the New Pocket Phantom with a Catphan Phantom  Calibration Study – Siemens Sensation 64 CT Scanner – 6 pocket phantoms placed in/near an anthropomorphic chest phantom – Catphan phantom also scanned – Varied slice thickness, mA, kVp, and pitch –  Pearson’s Correlation Coefficients – CT Density = 0.999 (P < 0.001) – Noise = 0.940 (P < 0.001) – Resolution = 0.929 (P < 0.001)  Open Source Lesion Sizing Toolkit The Lesion Sizing Toolkit http://public.kitware.com/LesionSizingKit/ The Lesion Sizing Toolkit (LST) is a free and open source software architecture designed to accelerate the development and evaluation of quantitative lesion sizing algorithms. Developed in 2008 Focused on Dissemination in 2009 RSNA Quantitative Reading Room of the Future Showcase Open Source Medical Imaging Software Course Benchmarks Volcano 2009 OSA ISP Special Issue on Imaging for Early Lung Cancer Detection Lung Cancer Risk Lung Cancer Formation • Significant tissue damage occurs as a result of particulate matter (PM) deposition  Hyaline Cartilage • Deposition is a function of air flow dynamics and PM characteristics • • Histology and CFD has shown up to a 100x greater PM deposition at: – Airway bifurcations  [Broday, Aerosol Science and Tech. 2004] – Respiratory bronchioles  [Churg & Brauer, Ultrastructural Path. 2000]  • Bifurcation and peripheral lung tissues likely exhibit some of the earliest preneoplastic changes in response to PM exposure Balashazy et al., J Appl Physiol 2003. Bifurcation Calcification in HRCT Bifurcation Calcification Open Image Archive 1.25mm Slice Thickness w/ Bone K Lung Cancer Risk Index (LCRI)   Features 1. Bifurcation Damage Index (BDI) • HRCT w/ B60f edge enhancing kernel • Mean of 5 airway bifurcations (~20min) • 2. FEV1/FVC • Decline associated with lung cancer risk • Follow ATS spirometry guidelines  BD CD  Classifier Method is Independent of Age, Gender, Pack Years… • Linear BDI vs. FEV1/FVC Hi gh er Hi gh er Regression line is for cancer cases scanned at 1mm slice thickness and FEV1/FVC > 55% Lo w er Lo we r Initial Performance Analysis Dataset Conditional Logistic Regression Cochran-Mantel-Haenszel (Odds ratio for a ∆ 0.033 in LCRI with (crude estimate) 1:3 matching ) 108 Cases Full Dataset OR = 1.84 CI: 1.18-2.85 p-value = 0.0067 OR = 2.89 CI: 1.02-8.19 p-value = 0.0467 67% sensitive 72% specific 79 Cases 1mm Only 100% sensitive 74% specific Conclusion: Individuals with higher LCRI are more likely to have lung cancer Data on 21 Cancers and 121 Controls COPD Pr ox im al Di st al Lu ng COPD Ca nc er Lung Cancer Risk Findings • Investigating a new quantitative imaging biomarker • Airway bifurcations are calcifying in a relationship with FEV1/FVC • In control cases, a significant trend observed between LCRI and age*pack years (P = 0.006) • Odds Ratio for LCRI is better than FEV1/FVC – LCRI = 2.73 (CI: 1.35-5.51, P = 0.005) = 0.44 (CI: 0.24-0.83, P = 0.005) – FEV1/FVC • Opportunities exist to identify new lung cancer patient sub-populations  Give-A-Scan Patient Donated Dataset Never Smoker, Cancer at 62, FEV1/FVC=84% Right 2.5mm Scan Standard Kernel Left Measuring Progress Interim Meetings 7 Workshops since 2004 1 Interim COPD Meeting Annual Workshop PCF/Cornell Database NCIA Give-A-Scan COPDGene? Large Open Image Databases QIBA FDA NIST Standards & FDA Approval Reproducibil ity & Comparison Early Clinical Trials BioChange & Volcano QIBA Studies Accelerate Developme nt of Therapy Assessmen t Methods Algorithms & Reference Methods Open Source Lesion Sizing Toolkit CT COPD Algorithms Publications Oncology Workshop Reports Quantitative CT Monograph ISP Oncology Special Issue Thank You Lung Cancer Risk Index Cancers and Age & PY Matched Controls (+/-10) PY = Age = 5 64 15 51 20 59 28 51 30 57 40 48 45 58 59 60 60 69 63 54 66 64 68 57 72 68 74 75 62 92 Cancer Subjects Sorted by Increasing Pack Years A control case was permitted to be used for more than 1 cancer case Lung Cancer Risk Index Cancers and Age & PY Matched Controls 1.0 mm CT Thickness Threshold 1.25 mm CT Thickness Threshold PY = Age = 5 64 15 51 20 59 28 51 30 57 40 48 45 58 59 60 60 69 63 54 66 64 68 57 72 68 74 75 62 92 Cancer Subjects Sorted by Increasing Pack Years A control case was permitted to be used for more than 1 cancer case (We are now using FEV1/FVC before bronchodilator) New Study Results (We are now using FEV1/FVC before bronchodilator) Thymom a New Study Results Carcinoid of the Thymus AAH
Comments
Top