Research Supervisor Connect
Automatic detection of normal mammography images
Medical Imaging and Radiation Sciences Research Group
PHD
The efficiency of breast screening programs relies on the accurate interpretation of normal mammograms. Currently normal cases make up approximately 98% of BreastScreen Australia's examinations contributing significant to radiologists' workloads and service expenses. At present, BreastScreen relies on paired reading of each mammogram even though there is a paucity of data on whether all normal mammograms require such a rigorous approach. It is well reported that increased fibroglandular tissue (density) can reduce specificity since obscuring radiopaque regions of the image may look suspicious. On the other hand a breast comprising of mainly fatty tissue is more radiolucent and may facilitate a greater level of confidence that the image is normal. Whilst several publications have provided some overall indicators for recognizing normal cases, there is little awareness of the specific features that define easy-to-interpret normal mammograms that may not require a rigorous double-reading approach. There is even less data supporting cost-effective solutions for the interpretation of the 800,000 normal cases that present each year in Australia.
The current work aims to build on the current evidence by examining a much larger series of normal mammograms available from the BREAST platform and BreastScreen NSW. In particular this work will have the following five aims:
1. Identify those cases where a second read did not add anything extra to the primary diagnosis;
2. Establish a descriptor containing the features that characterizes these easy-to-interpret cases;
3. Test our descriptor prospectively on a further group of cases that contain normal and cancer images;
4. Develop a computed aided platform to automatically recognise these normal images;
5. Identify the cost-effective implications of our recommendations.
With this project, there are opportunities here for PhD students interested in Diagnostic Imaging, Computer Science and Electrical Engineering.
The opportunity ID for this research opportunity is 2313