BME PhD Defense: Robert Ambrosini
Automated Detection and Growth Rate Determination of Metastatic Tumors in the Brain and Lungs
Supervised by Professor Walter OâDell
Whether they serve as the first indication of malignancy or the most recent sign of cancer progression within a patient's body, metastatic tumors must be detected early, diagnosed accurately, monitored carefully, and treated effectively in order to optimize patient outcomes with regards to both survival and quality of life. Advanced imaging techniques and newly developed therapy options make efficacious treatment of metastatic disease a possibility, however completely manual image reading of magnetic resonance (MR) and computed tomography (CT) scans hinders these efforts due to a lack of efficiency, the presence of subjectivity, and inaccuracies occurring as a result of operator fatigue. For this purpose, we have developed automated computer-aided diagnosis (CAD) systems to assist in detecting and sizing metastases in the brain and lungs.
In order to achieve CAD for metastatic brain tumors on MR scans, spherical tumor appearance models were created to match the expected geometry of brain metastases while accounting for partial volume effects and offsets due to the cut of MR sampling planes. A 3-D normalized cross-correlation coefficient (NCCC) was calculated between the brain volume and spherical templates of varying radii using a fast frequency domain algorithm to identify likely positions of brain metastases. Study results demonstrate that the 3-D template matching-based method can be an effective, fast, and accurate approach that could serve as a useful tool for assisting radiologists in providing earlier and more definitive diagnoses of metastases within the brain.
Automated brain lesion detection algorithms require prior brain extraction so as to lower computational cost and prevent high false positive rates. A method is developed that combines 3-D atlas matching with the image edge detection capabilities of active contours (snakes) for the purpose of performing brain extraction while preserving the MR appearance of lesions. Algorithm results are presented in the forms of similarity metrics as compared to gold standard extractions and analyses of lesion preservation and high probability false positive location removal.
The ability of a clinician to detect properly changes in the size of lung nodules over time is a vital element to both the diagnosis of malignant growths and the monitoring of the response of cancerous lesions to therapy. 3-D spherical template matching can be employed for the volume change determination of metastases in CT lung scans by identifying the best-fit template for each lung metastasis through the optimization of the 3-D NCCC calculated between the templates and the nodule. A total of 17 different lung metastases were extracted manually from real patient CT datasets, reconstructed in 3-D using spherical harmonics equations, and rescaled to multiple volumes to generate simulated nodule growth for algorithm testing. Results demonstrate the capacity of this approach to compute accurately small changes in volume at low computational cost.