Median Technologies · EURONEXT:ALMDT

Median Technologies — European Congress of Radiology 2022

11 slides from Median Technologies's conference presentation, Jul 15, 2022

  1. Slide 1: Development and Validation of a Machine Learning Based CADx in Lung Cancer Screening

    Development and Validation of a Machine Learning Based CADx Designed to Improve Patient Management in Lung Cancer Screening Programs Charles Voyton, iBiopsy® P. Baudot, C. M. Voyton, D. Francis, A. Baili-Laya, V. Bobin, B. Renoust, Y. Liu, A. Iannessi, B. Huet And the entire iBiopsy® team! Presentat

  2. Slide 2: Disclosures

    Disclosures All Authors are employed by: Median Technologies, Valbonne, France 2 Presentation at the European Congress of Radiology (ECR) 2022 | July 13-17, 2022 | Vienna, Austria

  3. Slide 3: Epidemiology

    Epidemiology Key points 1 • Mortality: Lung cancer is the leading cancer killer worldwide, accounting for 18% of all cancer deaths – A total of 1.8M deaths in 2020 (2.4M in 2030) – Low 5-year overall survival rate: 18% 2 • Incidence: 2nd most commonly diagnosed cancer representing 11.4% of all new c

  4. Slide 4: The High Value of LDCT Screening

    The High Value of LDCT Screening The NELSON & NLST studies results on LDCT screening Diagnosed outside of a screening programme Diagnosed in a screening programme › 20% reduction in mortality among patients who received LDCT vs X-RAY (and ~40% versus no screening) › A lung cancer detected by LDCT sc

  5. Slide 5: Unmet Medical Need – Issues With Implementation

    Unmet Medical Need – Issues With Implementation › Radiation exposure: Although the risk is low, there › Availability of quality testing: Important to be are still negative views against LDCT. screened at a center experienced in lung cancer screening and treatment. › False alarms: ~25% of LDCT screen

  6. Slide 6: Lung Screening Dataset – NLST trial

    Lung Screening Dataset – NLST trial NLST Dataset et CT-Arm 26,722 Selection for Download Patients • CT-Arm › Patients selected from the NLST NLST Downloaded Dataset • Cancer: All Screening Detected Cancer • Benign: all available with nodules and all imaging time points Dataset. Cancer Benign › 1,737

  7. Slide 7: CADx Classifier Model

    CADx Classifier Model Key features: Ensembling of very different deep Convolutional Neural Networks CADx › Input: Patch Containing Lesion › Output: Malignancy Prediction Per Patch 7 Presentation at the European Congress of Radiology (ECR) 2022 | July 13-17, 2022 | Vienna, Austria

  8. Slide 8: LCS Nodule Characterization Results (CADx)

    LCS Nodule Characterization Results (CADx) Automatic lung nodule characterization on all lung cancer stages › AUC of ROC on test = 0.99, max Youden Index Operating Point: Sensitivity = 95.3%, Specificity = 96.2% • Operating point: Sensitivity 98.6% at Specificity 80.2% A • Operating point: Sensitivi

  9. Slide 9: Translation in the wild

    Translation in the wild Enhancing Patient Recall Rate › NLST population (2.4% cancer prev), assuming › Recall/Non (< LR3 / > LR 3) the Lung-Rads performance (Pinsky 2015; 78% − Saved scans 57% decrease Sens, 94% Spec). − Faster scanned cancers (67% increase) › When used in a Recall/No-Recall paradig

  10. Slide 10: Conclusions and Takeaway Message

    Conclusions and Takeaway Message AI in Lung Cancer Screening › The model demonstrates robust performance for the characterization of pulmonary nodules detected by radiologists. ▪ 0.991 (95% CI: 0.987-0.995), with 98% sensitivity at a specificity of 90% › When deployed in a screening program, high pe

  11. Slide 11: Presentation at the European Congress of Radiology (ECR) 2022 | July 13-17, 2022 |

    11 Presentation at the European Congress of Radiology (ECR) 2022 | July 13-17, 2022 | Vienna, Austria

Median Technologies — European Congress of Radiology 2022 on 6ix