Author of

• 32851

  +

  PD01 - Poster Discussion Session (ID 4)

  • Event: NACLC 2019
  • Type: Poster Discussion Session
  • Track:
  • Presentations: 1
  • Moderators:
  • Coordinates: 10/11/2019, 15:45 - 16:45, Imperial Ballroom (B2 Level)

  • 33506

    +

    PD01.11 - Artificial Intelligence Can Detect Lung Cancer from High Resolution Microscopic Images of Conditioned Peripheral Blood (ID 127)

    15:45 - 16:45  |  Author(s): Cynthe Sims
    • Abstract

    Background:
    Lung cancer is a leading cause of death worldwide with about 2.1 million new cases and 1.8 million deaths expected during 2019. Screening for lung cancer however remains challenging. Low Dose Computed Tomography (LDCT) the approved screening test has low specificity and carries the risk of radiation. A non-invasive, specific and sensitive screening test is an urgent unmet public health imperative. Considering that pulmonary embolism is a significant risk in lung cancer, we hypothesized that detection of circulating emboli in peripheral blood using deep learning microscopy would be a credible approach to detect lung cancer.
    
    
    Method:
    We processed peripheral blood from 3977 asymptomatic individuals who underwent routine scans and multiple CA marker evaluation (1475 males [37 %], 2502 females [63%]), 89 patients of lung cancer (61 males [69 %] and 28 females [31%]) of whom 37 had detectable disease, 49 were on therapy and 3 had no radiological evidence of disease. Mono-nucleated cells obtained after centrifugation of blood samples were processed with CellWizardTM, a paradoxically cytotoxic cell media. Apoptosis resistant cells in the milieu are unaffected while cells with responsive cell death mechanism are killed. The process leaves behind Circulating Tumor Cells and Circulating Ensembles of Tumor Associated Cells (C-ETACs). High resolution images of the media wells holding the samples were then obtained on the 5th day. A deep machine learning algorithm was deployed with a training set of images from 44 samples each of asymptomatic individuals with category 1 and known cases of lung cancer respectively.
    
    
    Results:
    Among the 37 cases of lung cancer, the AI algorithm detected C-ETACs in 31 cases (84% sensitivity); 40 out of 49 (82%) patients with ongoing treatment could be detected. Patients with no evaluable disease were not classified as positive. Out of 3977 asymptomatic individuals, 2278 individuals were negative for all scans and CA markers; from whom 82 individuals (4%) were predicted by AI to be positive for lung cancer. AI evaluation of conditioned peripheral blood had sensitivity of 84% in detecting lung cancers, specificity of 96% in predicting patients negative for lung cancer with overall accuracy of 96%. The 82 individuals detected by AI as being positive for malignancy may have cancer other than that of the lung and are being monitored prospectively.
    
    
    Conclusion:
    High resolution microscopic images of conditioned peripheral blood coupled with an artificial intelligence algorithm is a cost-effective, non-invasive method to screen asymptomatic individuals for lung cancer without the risk of radiation.