INTERNATIONAL SOCIETY FOR ENDOMICROSCOPY (IS4E) PARIS, FRANCE NOVEMBER 27, 2017 Confocal Endomicroscopy in Urologic Surgery: urothelial carcinoma and other emerging applications Joseph C. Liao, MD Department of Urology Stanford University
Bladder cancer and UTUC Bladder cancer 5 th most common cancer in the U.S. 75-80% present as non-muscle invasive Ta/T1/TIS High recurrence rate up to 50-70% Upper tract urothelial carcinoma Shares histologic features with bladder ca 5-10% of all urothelial carcinoma Endoscopy integral part of management Tissue diagnosis and local staging Therapy: TUR and ablation Surveillance Jemal et al. CA Cancer J Clin, 2010. 60(5):277-300; Wood et al. Cambell-Walsh Urology, 10 th.ed. 713-25; Roupret et al. Eur Urol. 2013 Jun;63(6):1059-71.
Enhanced cystoscopy in current guidelines 2016 In a patient with NMIBC, a clinician should offer blue light cystoscopy at the time of TURBT, if available, to increase detection and decrease recurrence. (Moderate Recommendation; Evidence Strength: Grade B) In a patient with NMIBC, a clinician may consider use of NBI to increase detection and decrease recurrence. (Conditional Recommendation; Evidence Strength: Grade C) Hexaminolevulinate-assisted blue light cystoscopy (BLC) Narrow band imaging (NBI) Chang et al. 2016. https://www.auanet.org/education/guidelines/non-muscle-invasive-bladder-cancer.cfm
Confocal endomicroscopy Optical biopsy: dynamic in vivo confocal microscopy (488 nm) Approved for clinical use in Europe and U.S. (2015) Cellvizio, Mauna Kea Technologies Fluorescein as contrast agent Intravesical or intravenous route Probe-based (0.85 to 2.6 mm) Real time visualization of cancer microarchitecture and cellular features Sonn et al. J Urol. 2009. 182(4):1299-1305; Adams et al. J Endourol. 2011. 25(6):1-5; Wu et al. Urology, 2011. 78(1):225-31.
Endomicroscopy of bladder cancer optical imaging atlas Chen et al. (2014) Curr. Urol Rep. 15:437; Hsu et al. (2014) Cur Opin Urol. 24:66; Chang et al. (2013) J Endourol. 27(5):598-603
Interobserver Agreement TABLE 3. INTEROBSERVER AGREEMENT AND DIAGNOSTIC ACCURACY FOR CANCER DIAGNOSIS CLE WLC WLC + CLE Interobserver Agreement pa κ (95% CI) pa κ (95% CI) pa κ (95% CI) Experienced CLE urologists 90% 0.80 (0.45 1.00) 74% 0.46 (0.10 0.81) 90% 0.80 (0.45 1.00) Novice CLE urologists 77% 0.55 (0.46 0.64) 78% 0.54 (0.45 0.63) 80% 0.59 (0.50 0.68) Pathologists 81% 0.61 (0.47 0.76) - - - - Nonclinical researchers 77% 0.49 (0.38 0.60) - - - - Diagnostic Accuracy Sn Sp Sn Sp Sn Sp Experienced CLE urologists 84% 88% 89% 83% 89% 88% Novice CLE urologists 75% 81% 86% 79% 89% 83% Pathologists 84% 81% - - - - Nonclinical researchers 89% 73% - - - - pa = percent agreement; Sn = sensitivity; Sp = specificity Assess interobserver agreement and learning curve Urologists and nonurologists (n=10) 1-hour computer-based training n=32 patients Moderate agreement: κ= 0.55 Diagnostic accuracy of WLC + CLE Sensitivity 89% Specificity 88% Chang et al. J Endourol. 2013 May;27(5):598-603
Endomicroscopy of the resection bed following BLC-assisted TUR Chang et al. AUA 2017
Endomicroscopy - Upper Tract Bui et al. (2015) J. Endourol. 29:1418 Villa et al. (2016) J. Endourol. 30:237.
ENDOMICROSCOPY - UPPER TRACT n=14 IV fluorescein 0.85 mm probe URS manipulation not significantly impaired Imaging-pathological correlation Bui et al. (2015) J. Endourol. 29:1418
ENDOMICROSCOPY - UPPER TRACT Bui et al. (2015) J. Endourol. 29:1418
Endomicroscopy in Upper Tract n=11 topical fluorescein 0.85 mm probe in working channel of flexible URS Imaging followed by biopsy and laser ablation LG (7), HG (3), inflammation (1) Villa et al. (2016) J. Endourol. 30:237.
Goals: feasibility study develop imaging protocol description of imaging features n=21 in vivo/ex vivo imaging Prostatic and peri-prostatic structures Pathological correlation Vessel Adipocytes Nerve Lopez et al. (2016) J. Urol. 195:1110
Surgical margin assessment: extracapsular extension Extracapsular extension (pt3a, Gleason 3+4)
Endomicroscopy for small renal masses Optical diagnosis of SRMs Pilot ex vivo study to establish CLE imaging features Normal AML Benign Cancer n=20 clear cell Topical fluorescein/2.6mm probe oncocytoma papillary En face imaging -> pathological correlation cystic nephroma leiomyoma acquired cystic Su et al. (2016) J. Urol. 195:486.
Emerging areas M O L E C U L A R I M A G I N G C O M P U T E R - A I D E D I M A G E P R O C E S S I N G
Anti-CD47-FITC imaging with confocal endomicroscopy Pan et al., Sci Transl Med 2014;6:260ra148
Lurie et al. 2017 Biomed Opt Express 8:2106 Lurie et al. 2016 Biomed Opt Express 8:4995. 3D bladder reconstruction using standard WLC videos Longitudinal monitoring Optical annotation
Artificial Intelligence
Machine learning for image processing ARTIFICIAL INTELLIGENCE MACHINE LEARNING DEEP LEARNING CONVOLUTIONAL NEURAL NETWORKS (CNNs)
Convolutional Neural Network Dataset 81 subjects; 458 CLE videos 170,712 images 184 Cancer 70 Low grade papillary 91 High grade papillary 23 Carcinoma in situ 274 Benign 167 normal 107 inflammatory/other Collaboration with Stanford Biomedical Informatics GoogLeNet architecture 22 layers Training 80%, Validation 10%, Testing 10% Chang et al. EUS 2017 Szegedy C, et al. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2015:1-9
Results Diagnostic Accuracy of CLE 100% 80% 87% 79% 90% 60% 40% 20% 0% Accuracy Sensitivity Specificity Machine
Results Diagnostic Accuracy of CLE 100% 80% 87% 79% 79% 77% 90% 82% 60% 40% 20% 0% Accuracy Sensitivity Specificity Machine Urologists (n=8)
Real-time Feedback
High Grade Tumor
Summary confocal endomicroscopy in urology Platform technology that enables real time optical biopsy Bladder, upper tract, prostate, small renal masses Bladder and UTUC most promising Real time differentiation of cancer and benign, cancer grade (?) May improve yield of positive biopsy Combination with other enhanced endoscopy technologies Improve the quality of primary endoscopic intervention Current Limitations Single center experiences Going beyond initial proof-of-concept Clinical impact unknown Image interpretation Future Multi-center validation Molecular imaging Machine learning (convolutional neural network) for real time image processing
Acknowledgements Liao Lab Kathy Mach Tim Chang Bernhard Kiss Gautier Marcq Mandy Sin Ying Pan Dharti Trivedi Ying Pan Kristen Lurie Dimi Zlatev Timothy Chang Tommy Metzner Aristeo Lopez Mark Hsu Mandy Sin Emanuela Altobelli Michael Davenport Stephanie Chen Melanie Ngan Jen-Jane Liu Ruchika Mohan Lei Kang Collaborators Irv Weissman Eila Skinner Phil Beachy Sam Gambhir John Leppert Robert Rouse Audrey Ellerbee Zhen Cheng Funding NIH/NCI R01 CA 160986 Stanford Cancer Institute Urology Care Foundation NIH/NIAID U01 AI 082457 NIH/NIAID R44 AI 88756