Bayesian and grAphical Models for Biomedical Imaging

We introduce a graphical model for the joint segmentation and tracking of E. coli cells from time lapse videos. In our setup cells are grown in narrow columns (growth channels) in a so-called “Mother Machine” [1]. In these growth channels, cells are vertically aligned, grow and divide over time, and eventually leave the channel at the top. The model is built on a large set of cell segmentation hypotheses for each video frame that we extract from data using a novel parametric max-flow variation. Possible tracking assignments between segments across time, including cell identity mapping, cell division, and cell exit events are enumerated. Each such assignment is represented as a binary decision variable with unary costs based on image and object features of the involved segments. We find a cost-minimal and consistent solution by solving an integer linear program. We introduce a new and important type of constraint that ensures that cells exit the Mother Machine in the correct order. Our method finds a globally optimal tracking solution with an accuracy of > 95% (1.22 times the inter-observer error) and is on average 2 − 11 times faster than the microscope produces the raw data.

Dynamic contrast-enhanced MRI (DCE-MRI) images are increasingly used for assessing cancer treatment outcome. These time sequences are typically affected by motion, which causes significant errors in tracer kinetic model analysis. Current intra-sequence registration methods for contrast enhanced data either assume restricted transformations (e.g. translation) or employ continuous optimization, which is prone to local optima. In this work, we propose a new approach to DCE-MRI intra-sequence registration and pharmacokinetic modelling, which is formulated in an MRF optimization framework. The complete 4D graph corresponding to a DCE-MRI sequence is reduced to a concatenation of minimum spanning trees, which can be optimized more efficiently. To address the changes due to contrast, a data cost function which incorporates pharmacokinetic modelling information is formulated. The advantages of this method are demonstrated on 8 DCE-MRI image sequences of patients with advanced rectal tumours, presenting mild to severe motion.

The appearance of new Multiple Sclerosis (MS) lesions on MRI is usually followed by subsequent partial resolution, where portions of the newly formed lesion return to isointensity. This resolution is thought to be due mostly to reabsorption of edema, but may also reflect other reparatory processes such as remyelination. Automatic identification of resolving portions of new lesions can provide a marker of repair, allow for automated analysis of MS lesion dynamics, and, when coupled with a method for detection of new MS lesions, provide a tool for precisely measuring lesion change in serial MRI. We present a method for automatic detection of resolving MS lesion voxels in serial MRI using a Bayesian framework that incorporates models for MRI intensities, MRI intensity differences across scans, lesion size, relative position of voxels within a lesion, and time since lesion onset. We couple our method with an existing method for automatic detection of new MS lesions to provide an automated framework for measuring lesion change across serial scans of the same subject. We validate our framework by comparing to lesion volume change measurements derived from expert semi-manual lesion segmentations on clinical trial data consisting of 292 scans from 73 (54 treated, 19 untreated) subjects. Our automated framework shows a) a large improvement in segmentation consistency over time and b) an increased effect size as calculated from measured change in lesion volume for treated and untreated subjects.