Making the identification of criteria for further investigation in the diagnosis of breast cancer more effective
Breast cancer is the most frequent neoplasms in women worldwide, and despite continuous advances in diagnosis and treatment, it remains an important cause of death (1).
Consolidated scientific evidence has demonstrated the favourable effect of mammography screening on breast cancer mortality, thanks to the early and asymptomatic identification of tumour lesions (2,3). This makes it possible to treat small lesions in a less demolitive and more prognostically favourable manner. Mammography is therefore the cornerstone of breast cancer screening programmes and the first test to be performed in case of suspicious symptoms. At present, organised screening programmes are labour-intensive, considering the number of women who have to be screened for a cancerous lesion, and require the use of a double-reading X-ray system (5,6). Interpretation of mammographic examinations varies with the experience of the practitioner, and can be made difficult by the heterogeneous presentation of lesions and the masking effect of high-density breast tissue. In a population-based screening programme, the reading of mammographic examinations therefore requires the presence of a large number of dedicated radiologists, with a significant individual workload and critical systemic issues due to the lack of qualified readers (4,5).
Moreover, even in women who regularly participate in screening, some lesions become clinically manifest in the time interval between one screening mammogram and the next (false negative cases or interval cancers). Some of these lesions are retrospectively visible from previous screening mammograms. At the same time, some lesions may be sent for unnecessary further investigation because the mammogram may detect a suspicious lesion that is not confirmed by further investigations (false positive cases). Continuous efforts are being made in organised screening to reduce the probability of false positives and false negatives (6). In this context, the development of artificial intelligence systems is intended to improve the screening system and to support the radiologist in reading, on the one hand by maximising tumour detection and on the other hand by optimising the workload.
With this project, we plan to develop an artificial intelligence system based on screening mammography images to recognise women whose mammography test is clearly negative, that is, no lesions of any kind are detected. The objective is to bring the following benefits:
- Support the radiologist in reading the first level mammography screening examinations;
- Reduce reporting time and workload for screening radiologists;
- Speed up the identification of clearly negative women to be sent to the next screening round (at 1 or 2 years depending on age), so that readers can focus more on analysing suspicious images;
- Improving the efficiency of the system by increasing the already positive risk-benefit balance of screening programmes.
(1) Ferlay J, Colombet M, Soerjomataram I, et al. Estimating the global cancer incidence and mortality in 2018: GLOBOCAN sources and methods. Int J Cancer. 2019;144(8):1941-1953.
(2) Broeders M, Moss S, Nystro¨m L, et al. The impact of mammographic screening on breast cancer mortality in Europe: a review of observational studies. J Med Screen. 2012;19(suppl 1):14–25.
(3) Lauby-Secretan B, Scoccianti C, Loomis D, et al. Breast cancer screening–viewpoint of the IARC Working Group. N Engl J Med. 2015;372(24):2353–2358.
(4) Elmore JG, Jackson SL, Abraham L, et al. Variability in interpretive performance at screening mammography and associated with accuracy.Radiology 2009;253:641e51.
(5) Miglioretti DL, Smith-Bindman R, Abraham L, et al. Radiologist characteristics associated with interpretive performance
(6) Kopans DB. Digital breast tomosynthesis from concept to clinical care. Am J Roentgenol. 2014;202(2):299-308.