Earlier detection of breast cancer by computer assisted decision making in screening

Breast cancer

Breast cancer is the result of uncontrolled growth of breast cells. As with all cells in the human body, breast cells have a life cycle during which they grow, divide, produce more cells and eventually die. Under normal conditions, genes are able to effectively regulate this cellular life cycle. However, when genes develop an abnormality, the life cycle is affected and some cells may continue to divide even when new cells are not needed, forming a mass.

A tumor can be benign or malignant. Benign tumors are not considered cancerous: they grow slowly, and they do not invade nearby tissues or spread to other parts of the body. Malignant tumors are cancerous, and can spread beyond the original tumor to other parts of the body.

Breast carcinoma in a lymph node.jpg
Micrograph showing a lymph node invaded by ductal breast carcinoma. The dark purple (center) is lymphocytes (part of a normal lymph node). Surrounding the lymphocytes and extending into the surrounding fat (top of image) is ductal breast carcinoma. (source).

Breast cancer screening

The earlier breast cancer is found, the more likely it can be successfully treated. Therefore, the Dutch government has setup a nationwide screening for breast cancer for asymptomatic women between 50 and 75 years of age. Women are offered a X-ray examination of both breasts (mammogram) on a regular basis aimed at early detection of breast cancer. Introduction of the breast screening program has contributed to the reduction of the breast cancer mortality rate in the Netherlands. However, some cancers detectable on mammography images are missed. Traditional computer-aided detection (CAD) systems for mammography are intended to reduce false negatives by marking suspicious areas of the mammograms for radiologists to consider.

Mammograms.png
Mammograms taken at the biannual screening, with a lesion in the right breast. Left: medio-lateral oblique projection, right: cranio-caudal projection.

Purpose of the project

In breast cancer screening radiologists select a small percentage of women for referral based on suspicious abnormalities in their mammograms. To maintain high specificity, radiologists do not refer all abnormalities they see. Consequently, not all cancers initially detected are acted upon because of interpretation failure. We have found evidence that interpretation failure is a more common cause of missing cancers in screening than perceptual oversights. By detecting these missed cancers earlier the effect of screening can be increased. Computer aided detection (CAD) methods have been introduced as a way to avoid perception errors. This technology has gained wide acceptance in screening practice, in particular in the US. However, current CAD technology does not address the problem of interpretation failure in screening. In this project we will focus on this problem. The first goal of this project is to further develop our CAD schemes to optimize classification performance on regions identified as suspect by radiologists. The second goal is determine experimentally how CAD can be used as a decision aid in screening practice and to estimate to what extent this can increase earlier detection. In this project we will focus on detection of masses and architectural distortions, which are the most important patterns associated with invasive cancers.

Plan of investigation

The workplan consists of three parts: 1) Development of novel classification methods aimed at using CAD as a decision aid, 2) extension of existing databases with digital screening mammograms, and 3) an observer study in which the benefit of this approach is experimentally determined. In recent projects we conducted observer studies in which we collected more than 7000 localized mammographic findings of experienced radiologists, including full descriptions and malignancy ratings. This material will form the basis for investigation of new methods for decision support.

Relevance for cancer research

On the basis of recent data, it is estimated that due to screening breast cancer mortality in the Netherlands has decreased in the screened population by 800 cases per year. Methods developed in this project can lead to improved detection, less late stage detection, and further reduction of breast cancer mortality.

Results

Use of Contextual Information

When reading mammograms, radiologists do not only look at local properties of suspicious regions but also take into account more general contextual information. This suggests that context may be used to improve the performance of computer aided detection (CAD) of malignant masses in mammograms. We developed a set of context features that represent suspiciousness of normal tissue in the same case. For each candidate mass region, three normal reference areas were defined in the image at hand. Corresponding areas were also defined in the contralateral image and in different projections. Results show that the mean sensitivity in the interval of 0.05-0.5 false positives/image increased more than 6% when context features were added. This increase was significant (p<0.0001). Context computed using multiple views yielded a better performance than using a single view (mean sensitivity increase of 2.9%, p<0.0001). Besides the importance of using multiple views, results show that best CAD performance was obtained when multiple context features were combined that are based on different reference areas in the mammogram.

Feature Selection

Feature selection methods are often used to improve generalization performance of classifiers and shorten computation times. We investigated the effect of using a selection criterion that is similar to the final performance measure we are optimizing, namely the mean sensitivity of the system in a predefined range of the Free-response Receiver Operating Characteristic (FROC). It was found that significantly higher performances were obtained using feature sets selected by the general test statistic Wilks’ lambda than using feature sets selected by the more specific FROC measure. Features selection led to better performance when compared to a system in which all features were used.

CAD as a decision aid

An interactive computer-aided detection (CAD) workstation has been developed for reading mammograms to improve decision making. On this dedicated mammographic workstation (see figure below) the presence of CAD marks can be queried interactively by clicking on suspect regions in the mammogram using a pointing device. It is not possible to display all available CAD marks at once as in traditional CAD prompting devices. For each queried location, the workstation checks if a CAD mark is available at that location. If a CAD mark is available, it is presented to the reader by drawing displaying the contour of the region detected by CAD along with a computer-estimated malignancy score. The contour of the region is colored based on the malignancy score using a continuous color scale ranging from red to yellow, for respectively high to low malignancy ratings.

Cad decision aid maurice s.png

To evaluate the effectiveness of this novel concept, a reader study was conducted in which 4 screening radiologists and 5 non-radiologists with mammogram reading experience participated. The participants read 120 cases of which 40 cases had a malignant mass that was missed at the original screening. The readers read each mammogram both with and without CAD in separate sessions. Each reader reported localized findings and assigned a malignancy score per finding.

The mean sensitivity was 25.1% in the sessions without CAD and 34.8% in the CAD-assisted sessions. The increase in detection performance was significant (p=0.012). The mean sensitivity was computed in an interval of false positive fractions less than 10%. This interval was chosen because in screening programs radiologists usually have recall rates below 10 percent.

Lroc decision aid cad maurice.png

Average reading time was 84.7 ± 61.5 seconds/case in the unaided sessions and was not significantly higher when interactive CAD was used (85.9 ± 57.8 seconds/case).

Researchers

Funding

  • Program grant from the Dutch Cancer Foundation, number 2006-3655: Earlier detection of breast cancer by computer assisted decision making in screening (Project leaders: Nico Karssemeijer, Carla Boetes; time frame: 2006-2011)

Key publications

  • M. Samulski and N. Karssemeijer. "Optimizing Case-based Detection Performance in a Multiview CAD System for Mammography", IEEE Transactions on Medical Imaging 2011;30(4):1001-1009. Abstract/PDF DOI PMID

  • R. Hupse and N. Karssemeijer. "The effect of feature selection methods on computer-aided detection of masses in mammograms", Physics in Medicine and Biology 2010;55(10):2893-2904. Abstract/PDF DOI PMID

  • M. Samulski, R. Hupse, C. Boetes, R. Mus, G. den Heeten and N. Karssemeijer. "Using Computer Aided Detection in Mammography as a Decision Support", European Radiology 2010;20(10):2323-2330. Abstract/PDF DOI PMID

  • R. Hupse and N. Karssemeijer. "Use of normal tissue context in computer-aided detection of masses in mammograms", IEEE Transactions on Medical Imaging 2009;28(12):2033-2041. Abstract/PDF DOI PMID

  • R. Hupse and N. Karssemeijer. "The use of contextual information for computer aided detection of masses in mammograms", in: Medical Imaging, volume 7260 of Proceedings of the SPIE, 2009, page 72600Q. Abstract/PDF DOI

  • M. Samulski, A. Hupse, C. Boetes, G. den Heeten and N. Karssemeijer. "Analysis of probed regions in an interactive CAD system for the detection of masses in mammograms", in: Medical Imaging, volume 7263 of Proceedings of the SPIE, 2009, page 1, page 726314. Abstract/PDF DOI