Breast cancer is the most frequent cause of cancer-related death in women worldwide. The past decades have seen a steady increase in cancer incidence, but luckily not in mortality due to improved treatment and screening. In order to further decrease the mortality rate it is paramount to detect breast cancer in an as early stage as possible.
To aid in the early detection of signs of breast cancer, many countries have implemented breast cancer screening programs in which women are at a regular interval invited for an examination where X-ray images of the breast are made, the so-called mammograms. In such an examination each breast is compressed and imaged in two different directions. These images are, perhaps together with earlier scans, assessed by expert radiologists for signs of breast cancer.
In the BI-RADS classification there are roughly speaking two large groups of such possible signs: masses and calcifications. Masses are a dense group of cells which grow in an abnormal, and sometimes uncontrolled way. Their shape and the degree to which the boundary is well-defined is an important factor in assessing if the mass is malignant or not.
On the other hand, calcifications appear as very small bright dots on the mammograms and are typically less than 0.5mm in diameter. They are very common as they can be caused by several natural processes such as calcification of the arteries, aging or injuries and are usually benign. Malignant calcifications often appear in clusters, which can be an important sign of certain cancers which are not yet invasive, such as the ductal carcinoma in situ (DCIS). Early detection is important, so that appropriate action can be taken before these clusters become invasive lesions.
In the national screening programs, only a small proportion of the acquired examinations will be suspicious and will warrant further investigations. As the signs are often subtle and relatively rare amongst all the normal scans, a computed aided diagnosis (CAD) program can be used to improve the detection rate, because radiologists sometimes miss these subtle microcalcifications. In this project will develop CAD algorithms using machine learning techniques for the automatic screening of mammograms.
A CAD pipeline can be split in two parts, one for the masses, and one for the calcifications. The state-of-the art for detection and classification of masses has been recently published by the DIAG group .
The goal of this project is to automatically detect and classify individual or clusters of calcifications as benign/malignant using machine learning techniques. There has been some progress lately on the detection of individual calcifications [1, 2] and the related false positive reduction algorithm and therefore the project will focus on the classification of clusters of calcifications and/or individual calcifications.
Depending on the length of the project and the interests of the student we can apply different deep learning techniques, such as the conventional convolutional neural networks, but also more advanced techniques such as regional CNNs  and reinforcement learning .
This project is of particular interest to students who want to experiment and gain experience with state-of-the-art neural network models on a large medical dataset.
1. Mordang, J.-J., Gubern-Mérida, A., den Heeten, G., & Karssemeijer, N. (2016). Reducing false positives of microcalcification detection systems by removal of breast arterial calcifications. Medical Physics, 43(4), 1676. https://doi.org/10.1118/1.4943376
2. Mordang, Jan-Jurre, Janssen Tim, Bria Alessandro, Kooi Thijs, Gubern-Mérida Albert, K. N. (2016). Automatic Microcalcification Detection in Multi-vendor Mammography Using Convolutional Neural Networks. In Tingberg A., Lång K., Timberg P. (eds) Breast Imaging.
3. Kooi, T., Litjens, G., van Ginneken, B., Gubern-Mérida, A., Sánchez, C. I., Mann, R., … Karssemeijer, N. (2017). Large scale deep learning for computer aided detection of mammographic lesions. Medical Image Analysis, 35, 303–312. https://doi.org/10.1016/j.media.2016.07.007
4. Maicas, G., Carneiro G., Bradley, A., Nascimento, J., Reed I., (2017). Deep Reinforcement Learning for Active Breast Lesion Detection from DCE-MRI. "MICCAI 2017 conference proceedings Part III, pp. 665"
5. Girshick, Ross. (2015). Fast r-cnn. "Proceedings of the IEEE international conference on computer vision."
The work will be executed in a high level environment for medical image analysis, at the Diagnostic Image Analysis Group (DIAG) of the Radboud University. DIAG is a leading research group in computer-aided detection and diagnosis (CAD).