Clinical Trials in Radiology
08:58V. Dahlblom, Malmu00f6 / SE
To investigate whether additional digital breast tomosynthesis (DBT) screen-detected cancers could be detected on the corresponding digital mammography (DM) examination by a deep-learning based AI system, using paired data from a prospective tomosynthesis screening trial.Methods and materials:
We used a subgroup of 14768 women from the Malmö Breast Tomosynthesis Screening Trial. All participants were examined with both DM and DBT with independent readings. Of 136 screen-detected cancers, 41 cancers were detected only on DBT. We analysed DM with a mammography AI software (ScreenPoint Transpara™). The software identifies and scores suspicious areas in the images with a score between 1 and 100 (highest risk), and also provides a composite cancer risk score between 1 and 10 (highest risk) for the whole examination. A cancer was defined as detected by the software if the examination got the highest risk score and the cancer lesion was correctly localised and scored over 60. Descriptive statistics were used.Results:
1797 examinations (12%) got the highest examination risk score. The AI software detected 15 of the 41 DBT-only detected cancers (37 %) on DM. A majority were invasive cancers, with 8 invasive ductal carcinomas, 4 invasive lobular carcinomas, and 1 invasive tubular carcinoma. Three cancers had histological grade 3, and 2 had lymph node metastases.Conclusion:
A substantial number of the additional DBT screen-detected cancers were also detected on DM with AI. The possibilities of using mammography plus AI in screening should be further investigated, taking effects on e.g. false positive recalls into account.Limitations:
Single-centre study with a single vendor.Ethics committee approval
Approved, informed consent used.Funding:
AIDA Fellowship, Vinnova (V. Dahlblom) and Governmental Funding for Clinical Research, Region Skåne. ScreenPoint Medical provided the software, but did not participate in the study.
07:13B. den Dekker, Utrecht / NL
To identify patient and MRI characteristics that are useful to distinguish true-positive screening MRI examinations from false-positives in women with extremely dense breasts.Methods and materials:
Patient and MRI characteristics of 454 Dutch breast cancer screening participants (50-75 years) with Volpara density category 4 on mammography and a positive first-round MRI screening result (BI-RADS 3/4/5) after a normal screening mammography, were collected prospectively (DENSE trial). Positive predictive values for malignancy in relation to patient and MRI characteristics were assessed. Multivariable logistic regression analysis with stepwise backward selection using AIC was used to find the optimal model to distinguish true-positives from false-positives.Results:
Of 454 women with a positive MRI screening examination, 79 were diagnosed with breast cancer, the remaining 375 MRI examinations were considered false-positive. In univariate analysis age (p-value<0.03), BMI (p-value 0.01), first degree relatives with breast cancer (p-value 0.002), MRI BI-RADS (p-value<0.001), lesion type (p-value 0.02) and lesion kinetics (p-value 0.02) were associated with breast cancer diagnosis. The optimal multivariable prediction model included MRI BI-RADS, age, previous breast biopsy, first degree relatives with breast cancer, and menopausal status (AUROC 0.829, 95%CI 0.779-0.880).Conclusion:
In addition to BI-RADS classification of screening MRI, the patient characteristics of age, previous breast biopsy, first degree relatives with breast cancer, and menopausal status might be useful to distinguish true-positive MRI examinations from false-positives in women with extremely dense breasts.Limitations:
The model has not yet been validated in an independent dataset.Ethics committee approval
Ethical approval for the DENSE trial was obtained from the Dutch Minister of Health, Welfare and Sport (2011/19). Written informed consent from study participants was obtained.Funding:
Supported by UMC Utrecht, ZonMw, Dutch Cancer Society, Dutch Pink Ribbon/A Sister’s Hope, Bayer AG Pharmaceuticals, stichting Kankerpreventie Midden-West, Volpara Health Technologies.
09:15D. Dabli, Nu00eemes / FR
The French Society of Medical Physics (SFPM) initiated a task group to establish dose reference levels (DRL) for the most common procedures performed in operating rooms using mobile X-ray systems.Methods and materials:
This multicenter prospective study involved 57 french medical institutions of different categories. Anonymised data was collected for 10 to 30 consecutive interventions from a list of 62 procedures types, belonging to 7 surgery specialties (neurosurgery, orthopaedics, digestive, urology, cardiology, vascular, and multi-specialty). Dose-Area-Product (DAP) and fluoroscopy time Reference Levels (RL) were established based on the 3rd quartile of distributions. Correction coefficients from quality assurance reports were applied to DAP displayed values to improve their accuracy. The impact of the procedure complexity was also investigated.Results:
DRLs were established for 31 procedures performed with mobile C-arms. Important variations were observed between the different surgery specialties. Lowest doses were reported for orthopaedic procedures (DAP=0.07 Gy.cm2 for hallux valgus, elbow, and hands), while highest doses were found in vascular procedures (DAP=85 Gy.cm2 for abdominal aortic aneurysm repair). In urology, DRLs ranged between 2 Gy.cm2 for ureteroscopy and 10 Gy.cm2 for lithotripsy. In cardiology, DRLs were higher for 3-leads pacemakers than for 1 or 2-leads pacemakers.
Moreover, results show values below those obtained from fixed interventional radiology facilities. Indeed, when compared to the french RL using fixed modalities, DAP is 6 times lower for lower-extremity arteriography and 3 times lower for biliary drainage.
DRLs were established to help medical physicists and surgeons to evaluate their practice and optimise patient radiation safety.Limitations:
Patients' body mass index was not considered.Ethics committee approval
This study was approved by the National Commission for Data Protection and Liberties and registered in Clinicaltrials.gov.Funding:
No funding has been obtained.
07:36M. Vonder, Groningen / NL
To determine and optimise the diagnostic performance of the Netherlands-China Big-3 screening (NELCIN-B3) protocol for early detection of lung cancer, cardiovascular disease (CVD), and COPD by low-dose CT in a Chinese population.Methods and materials:
Diagnostic randomised controlled and prospective cohort studies are conducted in three hospitals in China. 14,000 participants between 40-74 y will be randomly allocated to intervention and control groups. Risk-factor questionnaires, lung function, and blood samples will be collected. Intervention group participants will undergo low-dose chest CT and management according to NELCIN-B3 protocol. Quantitative assessment of Big-3 imaging biomarkers will be performed: lung nodule volume, volume doubling time, coronary calcium score, and emphysema score. Control group participants will undergo default chest CT and be managed according to hospital protocol based on lung nodule diameter and visual assessment of COPD and coronary calcium. Four years after the initial assessment, the referral rate, incidence of the Big-3, and related mortality will be evaluated.Results:
The effectiveness of quantitative assessment of CT imaging biomarkers for the Big-3 in the intervention group will be evaluated and compared to the control group with regards to referral rate, clinical diagnosis of the Big-3, and related mortality.Conclusion:
We expect that the quantitative assessment of the CT imaging biomarkers will reduce the number of unnecessary referrals for early detected lung nodules and improve the early detection of CVD and COPD in a Chinese population.Limitations:
Management of non-solid nodules in the intervention group will be based on the diameter due to lack of validated volume-based protocol.Ethics committee approval
Approval was issued by the Biomedical Research Ethics Committee of Shanghai Changzheng Hospital and written informed consent collected.Funding:
This study is funded by the Chinese Ministery of Sciences and Technology (MOST) and the Royal Dutch Academy of Sciences (KNAW).