The second group of further work focuses on the development of new research directions stemming from this thesis. We think that these can be classified again in four different categories: looking for masses in the pectoral muscle, looking for other types of lesions, diagnosis of the found masses, and, finally, breast density quantification.
We have focused this work on looking for masses only in the breast regions. However there are some marginal cases where masses also appear in the pectoral muscle. We actually think that our approach would be useful also for finding this kind of masses, although we have not tested this. This assumption relies on the fact that the pectoral muscle appearance is more uniform than the breast tissue. A major problem would be those situations were the mass is located in the pectoral-breast boundary, as its neighbourhood will have very different grey-level values.
Another research direction will be the application of the developed algorithm to the detection of other types of mammographic lesions, like micro-calcifications or spiculated lesions. At a first glance, both the template approach and the false positive reduction methods could be easily applied for both kinds of diseases. However, more investigation has to be done in order to validate those assumptions.
The third aspect is the diagnosis of the found masses, which clearly is distinct from both explained further works above, where we only talk about detection. Once the masses have been found they can be diagnosed. For such a task, the shape and margin of the lesion have to be accurately studied. We have shown that our algorithm extracts a ``rough" information of the mass, including the centre of masses and the lesion size. The traditional way to capture the shape of the lesion starting from the obtained bounding box would be the use of a snake, like the ones independently proposed by Kobatake et al. [89] and Sahiner et al. [156,158]. However, we think that our template algorithm might be capable of adapting correctly to the mass if it is initialized using an accurate manual segmentation.
Finally, the fourth future topic is the breast density quantification using not only X-ray information but also using Magnetic Resonance Imaging (MRI). The key point here is to use the segmentation strategy not only in mammographic images but also in MRI.