Accuracy of Breast Density Estimation from Mammographic Images

N. Geeraert (1, 2, 3, 4) , Rémy Klausz (2) , Isabelle Bloch (3, 4) , S. Muller (2) , H. Bosmans (1)

1 KU Leuven - Catholic University of Leuven = Katholieke Universiteit Leuven
2 GE Healthcare - General Electric Medical Systems [Buc]
3 IMAGES - Image, Modélisation, Analyse, GEométrie, Synthèse
4 TSI - Département Traitement du Signal et des Images

Breast density has been defined as an important risk factor for the development of breast cancer but the mechanisms of the impact on breast cancer development remain unsolved. One of the main discussions is the definition of breast density. Traditionally breast density is derived by dividing the area of the fibroglandular tissue in the image by the area of the total breast. From a physics point of view the ratio of volumes is a much more representative measure of the dense tissue in the breast. We evaluate the sensitivity of a method that computes the ratio of the volume of the fibroglandular tissue by the volume of the total breast from mammography projection images via a calibration procedure of the imaging system with breast phantom material and acquisition parameters of the image. The method was tested on phantoms with known density and thickness. We compared the results of our method to the literature. Two different authors described the errors on their model for density measurements on phantoms. Highnam et al. found an average error of 1.11% for 5 times 5 plugged densities below 38% density and Pawluczyk, Yaffe et al. found an error below 5% for inserted phantom disks (1 cm, 100%, 75%, 50%, 25%, 0%), on digitized screen-film images. The density model that we developed to compute the density estimation for phantoms has an accuracy equivalent to the best results in literature. The next step will be to improve the control of the influence of the breast thickness estimation in order to control the error of the density estimation for real breasts.

Mots clés: mammography; breast density; system calibration

Domaines: Intelligence artificielle [cs.AI]; Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]; Traitement des images [eess.IV]; Robotique [cs.RO]

N. Geeraert, Rémy Klausz, Isabelle Bloch, S. Muller, H. Bosmans. Accuracy of Breast Density Estimation from Mammographic Images. BHPA, Feb 2013, Mechelen, Belgium. pp.4. ⟨hal-00843356⟩