Mobiele menu

Breast CAlcification Risk Evaluation Study (Breast-CARE Study)

Projectomschrijving

Kalkspatjes gevonden bij bevolkingsonderzoek op borstkanker: risicovol of niet?

Doel

Jaarlijks worden door het bevolkingsonderzoek op borstkanker zo’n 24.000 vrouwen naar het ziekenhuis verwezen voor nader onderzoek. Een derde daarvan heeft als enige afwijking kalkspatjes in de borst die kunnen wijzen op borstkanker of op de voorloper van borstkanker ‘Ductaal Carcinoom In Situ’ (DCIS). Bij de meerderheid van deze vrouwen is daar echter geen sprake van. Zij zijn dus achteraf onnodig verwezen, vaak met onrust en mogelijk ook overdiagnostiek en overbehandeling tot gevolg. Dat willen we in de toekomst voorkomen.

Aanpak en verwacht resultaat

In het Breast-CARE project zullen we de radiologische, pathologische en moleculaire gegevens van de gevonden afwijkingen met kalkspatjes combineren om met kunstmatige intelligentie te leren welke wel en welke niet kunnen wijzen op borstkanker of DCIS. We hopen zo uiteindelijk vele vrouwen de last van achteraf onnodige verwijzingen te besparen.

Verslagen


Samenvatting van de aanvraag

BACKGROUND Calcifications as detected by mammographic screening might indicate the presence of breast cancer. Therefore, women with suspicious screen-detected calcifications are referred to the hospital for further diagnostic evaluation causing about one third, i.e. 8,000 women annually, of all referrals in the Netherlands. Only for a minority (~700 women/year), these calcifications are related to invasive breast cancer (IBC). For about 1,800 women per year, these represent the breast cancer precursor Ductal Carcinoma In Situ (DCIS). Consequently, 5,500 women annually have ‘harmless’ calcifications and do not need referral. Obviously, it would be a major gain to learn how to distinguish high- from low-risk calcifications to spare many women unnecessary referrals. OBJECTIVE Breast-CARE (‘Breast CAlcification Risk Evaluation’) aims to avoid unnecessary referrals of screen-detected calcifications by applying artificial intelligence (AI) on screening mammograms by distinguishing high-risk from low-risk calcifications, as the latter group do not need referral. PLAN OF INVESTIGATION To achieve our aim, we will take the following steps: 1. Compiling an unbiased cohort of women with screen-detected calcifications. We will start with the cohort of BOZW (‘Bevolkingsonderzoek Zuid-West Nederland’) of ~10,000 women referred for pure calcifications between 2010 and February 2018. 2. Risk assessment of the presence or development of ipsilateral DCIS or IBC after referral of screen-detected calcifications. We will link the cohort as described above to data from the NCR (Netherlands Cancer Registry) and PALGA (National Pathology Archive) to analyse the pathology at the time of primary diagnosis related to mammography results. Additionally, we will assess the risk of developing subsequent ipsilateral DCIS or IBC of both treated and untreated women referred for calcifications, as we will evaluate the follow-up of the women included at the end of this study. 3. Developing an AI-based risk-score of malignancy for mammographically detected calcifications. To develop such a model, we will extend work currently done in our concurrent KWF IMAGINE project. For the task of developing an AI-based risk-score, we proceed in two steps. First, we will apply a u-net based architecture to segment the suspect calcified lesion. Second, we will classify this lesion with a deep neural network to obtain an AI-based malignancy score (AMS). We will also explore the use of classical feature analysis method (`radiomics’) as a baseline model. 4. In-depth pathological analysis of calcification-associated lesions and correlate this with the AI-based malignancy score (AMS). Correlating the AI-score with pathological features aims to provide new insights which pathological and molecular features ‘explain’ the AMS. To do so, we will request about 1,440 pure, primary calcification samples via PALGA of a large series of pure, primary calcifications. 5. Evaluation if AMS helps screening radiologists in calcification risk assessment. To validate our algorithm, we will perform a multireader, multicase study. In this study at least 6 radiologists will read the images in the study either with or without AMS support. The same process will be repeated three months later to allow for a proper comparison where each radiologist has read each image both with and without algorithm, being blinded for their previous choice. 6. Validating the algorithm in independent cohort. As an independent validation cohort, the PREVENTICON dataset, the screening cohort from BOMW (‘Bevolkingsonderzoek Midden-West Nederland’), will be used. The PREVENTICON dataset has a similar size as the data available through BOZW. As there is only negligible overlap between women screened in both regions, this can be regarded as a complete independent validation. COMPLIANCE TO PRIVACY REGULATIONS This proposal has been discussed with the director of the population-based screening South-West (BOZW) and the National Institute of Health and Environment (RIVM). We made a start to arrange all data and material transfer agreements between the relevant stakeholders to secure this research will be done compliant to the privacy regulations. RELEVANCE Ultimately, the AMS will innovate and tailor screening, saving many women the burden of unnecessary recalls, overdiagnosis and overtreatment. This will also aid to limit cancer worries and anxiety, needless hospitalisation, and reduce health care costs. NEW KNOWLEDGE The AMS will help the screening radiologists to distinguish higher risk screen-detected calcification-associated lesions of the breast that do need referral to the hospital for further diagnostic evaluation from those that do not need this. IMPLEMENTATION If the AMS can reduce the referral rate safely and significantly, we will take steps in collaboration with the relevant stakeholders (RIVM, Dutch Expert Center for Screening, screening radiologists) to implement this in routine breast cancer screening practice.

Kenmerken

Projectnummer:
555004201
Looptijd: 75%
Looptijd: 75 %
2020
2025
Onderdeel van programma:
Gerelateerde subsidieronde:
Projectleider en penvoerder:
dr. J. Wesseling MD PhD
Verantwoordelijke organisatie:
Nederlands Kanker Instituut