Author of

• 20149

  +

  MA 20 - Recent Advances in Pulmonology/Endoscopy (ID 685)

  • Event: WCLC 2017
  • Type: Mini Oral
  • Track: Pulmonology/Endoscopy
  • Presentations: 1
  • Moderators:C. Lee, S. Sasada
  • Coordinates: 10/18/2017, 14:30 - 16:15, F205 + F206 (Annex Hall)

  • 24386

    +

    MA 20.08 - Classification of Confocal Endomicroscopy Patterns for Diagnosis of Lung Cancer (ID 9874)

    15:15 - 15:20  |  Author(s): O. Ramos
    • Abstract
    • Presentation
    • Slides

    Background:
    Solitary pulmonary nodules diagnosis and management is so challenging that nNew endoscopic techniques are being introduced to reduce uncertainty in peripheral pulmonary lesions (PPL) diagnosis and management. increase its diagnostic yield. Probe-based confocal laser endomicroscopy (pCLE) is a technique that can microscopically image the lung tissue in vivo during flexible bronchoscopy, though it can be difficult for pulmonologists to distinguish cellular patterns in a monochrome vision under respiratory and cardiac movements. . The goal of this work is to explore explore if Computed-Aided Diagnoses (CAD) tools can obtain a reliable diagnoses with pCLE in lung cancer.
    
    Method:
    A pilot study using 2 different methods for pCLE pattern analysis was performed:, one based on visual analysis by 3 experts and the other one based on computeron computerized analysis of visual patterns called Graphcom. Twelve 12 pCLE videos ( obtainedobtained using mMethylene blue dye (1%) and Alveloflex-Cellvizio 660nm miniprobe) were selected from patients with endobronchial lesionsperipheral SPNs (6 with lung adenocarcinoma cancer and 6 with inflammatory disease) during rigid bronchoscopy under general anesthesia. Afterwards, Vvideo sequences from pCLE were visually explored by one of the authors to select between 10 and 15 framesimages that presented a clear cellular pattern, without artifacts. . These images were shown to 3 observers who were familiar with confocal images but ignored the final histopathological diagnosis for a blind visual labellinglabeling. Images were also computationally analyzed using methods from social networks community analysis in a graph representation of pCLE images based on visual features to potentially overlapping groups of images that share common visual properties.
    
    Result:
    Our preliminary results indicate that on average visual analysis with 3 independent experts can only achieve a 60.2% of accuracy and has large variability amongst observers, while the accuracy of the proposed unsupervised image pattern classification rai(GraphCom) sesrises to 83,4.4%.
    
    Conclusion:
    Visual inspection of CLE images from lung tissue fails to provide accurate diagnosis. CLE images contain enough visual information for in vivo detection of neoplastic cell patterns that could be discriminated using cComputation methods and graph structural analysis applied to deep-learning feature spaces can increase diagnostic accuracy of pCLE images against visual analysis (83.4% vs 60.2%). Future studies are needed to apply this method in a real time scenario during bronchoscopy for PPL diagnoses.
    
    

    Only Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login, select "Add to Cart" and proceed to checkout. If you would like to become a member of IASLC, please click here.

    Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.