Computer Aided Detection and Diagnosis for MRI Breast Screening: MRI is now established as the most effective screening modality for high risk groups of women. There are two main challenges to overcome with the reporting of MRI screening mammography - the false positive rate is high leading to many unnecessary biopsies and the time taken to report each exam is high which makes the provision of MRI screening an expensive proposition to a health care provider. Computer Aided Detection and dignosis systems can play a role in ameliorating both of these problems. We have demonstrated that it is possible to acheive high accuracy using features based on the DCE data (Levman et al. 2008) and, more recently on textural and morphological features (Gallego-Ortiz & Martel 2015). We are also developing a fully automated system which includes fast motion correction (Martel et al. 2007), automatic segmentation of the chest wall (Martel & Gallego 2013; Khalvati et al. 2014) as well as the detection of suspicious lesions using neural networks.
Digital Pathology: Histopathology is the "gold standard" against which other imaging techniques are validated and it also provides a wealth of prognostic and predictive information that can be used to guide therapy and understand disease. Digital pathology allows image processing techniques to be applied to microscopy images. The Martel Lab has developed Machine Learning techniques for the analysis of wholemount breast H&E images (Peikari et al. 2015). We have also developed a pipeline for reconstructing 3D pathology image volumes from serial sections that does not require implanted fiducials, can work on breast whole mount images and that has the potential to be useful for sparsely sampled data (Shojaii et al. 2014). A start-up company, Pathcore Inc., was spun out of the lab to further develop and improve on viewing and analysis software for whole slide imaging and to build a management platform for digital pathology data. The Martel lab, Pathcore and several US labs are now collaborating on an NIH NCI project to develop a pathology image informatics platform (PiiP) which will facilitate the development, sharing, validation and clinical translation of software for the analysis of wholeslide images
Quantitative Analysis of Dynamic Contrast Enhanced Images: Functional in vivo information can be extracted from dynamic contrast-enhanced (DCE) sequences of images and there are many applications in cancer imaging both in diagnosis and in the assessment of therapeutic response. There are, however, many challenges in ensuring that analysis methods are reproducible and that the contributions to the changing signal intensity due to different tissue components can be separated. The Martel lab has developed various unmixing methods based on PCA, Factor analysis and, more recently ICA (Mehrabian et al. 2013). These methods have been applied to a wide range of imaging modalities including dual-photon microscopy (Mehrabian, Lindvere, et al. 2012) and contrast-enhanced ultrasound (Lueck et al. 2008). Using these approaches we are able to to separate intravascular and extravascular components from DCE-MRI images (Mehrabian, Chandrana, et al. 2012) allowing for more accurate modelling of tumour pharmacokinetics (Mehrabian et al. 2015).
Image Registration for DCE-MRI: In order to peform analysis on DCE image sequences it is first necessary to correct for motion. Image registration is particularly challanging in contrast-enhanced studies as it is necessary to differentiate beteen patient motion and the signal intensity changes of interest. We developed and validated a method of correcting for deformable motion in DCE-MRI breast studies in less than 2 minutes/volume. This technique resulted in a significant improvement in image quality. The resulting software was licensed to Sentinelle Medical who specialize in the manufacture of dedicated MR breast coils and MR guided biopsy guidance software (they are now part of Hologic) (Martel et al. 2007). The work has therefore been translated effectively into the clinical domain as it is now incorporated into the workflow of radiologists using the Aegis software. We have also worked on motion correction in the abdomen - a more challenging problem due to the large respiratory motion and increase contrast enhancement due to the presence of large vessels and highly perfused organs.