The CIA Lab has developed/is developing a series of computer software to advance quantitative radiology. The demonstration videos and images of the technologies can be found at the following website: http://www.columbia.edu/~bz2166/Technologies.html The lesion and organ contours were generated strictly by our software, with no human correction.
This algorithm combines the image analysis techniques of region-based active contours and level set approach. The algorithm requires manual initialization by drawing a lesion region-of-interest on one single image and then it automatically segments the entire lesion volume. The software is easy-to-use, fast, and robust. It can be applied to segment both homogeneous and heterogeneous lesions such as liver lesions (HCC) on multiple phase CT and MRI and brain lesions on MRI. The lesions can be hyper- and hypo-density compared to the surrounding parenchyma.
With the increased significance of medical images in diagnosis, treatment, and drug evaluation, the issue of image quality deserves more attention. In clinical practice, incomplete CT scans (i.e., missing body-part) and inappropriate contrasted scans result in recall of patients and unnecessary scans. In multicenter clinical trials, images acquired from different centers are sent to the Principal Investigator’s institution or to centralized data management laboratories for storage and subsequent analyses. Compliance of the received imaging data with study-defined (imaging) protocol is crucial for the success of a trial. This technology can help automatically identify non-compliant study scans in terms of the coverage of specific body parts (i.e., the lungs, abdomen and pelvis) and contrast enhancement based on analysis of image content.
This technology proposes an algorithm combining marker-controlled watershed and geometric active contours in a unique way to segment volumes of both small nodules and large masses in the lungs with highly satisfactory accuracy, reproducibility and speed. The algorithm requires manual initialization of a region-of-interest that encloses the lesion to be segmented on a single image and then it obtains the entire tumor volume automatically. The advantages of this algorithm lie in its abilities to separate lung lesions attached to anatomical structures with densities that are the same or similar to that of the lesions and segment part-solid and ground glass opacity (GGO) nodules.
This technology presents a two-stage segmentation method, i.e., sphere subdivision followed by an active contour method. The technology is computationally efficient and enables accurate volumetric measurement for lymph nodes.
These technologies can segment single object volume (e.g., liver, spleen, kidney) as well as simultaneously segment multiple object volumes (e.g., tumors and organs). It incorporates two established methods (the region-based active contour model and watershed algorithm) into a unified framework and provides greater accuracy in the separation of multiple objects that are physically connected to each other (e.g., liver and kidney, liver and heart, spleen and kidney).
This technology describes an algorithm and graphical user interface-based software for efficiently and accurately separating liver right and left lobes in images generated by MRI, CT, and other medical imaging technologies. Currently available technologies achieve lobe separation either by manual drawing on each slice (which is time consuming and tedious) or by finding markers on the surface of the liver (which yields poor accuracy). This technology relies on more accurate landmarks within the liver anatomy and an algorithm to automatically segment the liver into its left and right lobes based on the landmarks. The landmarks can be obtained either semi-automatically or automatically and yield real-time separation of the lobes. Refinement of the separation result can be achieved using the graphical user interface in a real-time fashion. Applications of the technology include 1) radiation and surgical treatment planning and 2) patient selection and follow-up checkups for liver transplantations.
Monitoring of lymph nodes is a common approach for detection of cancer metastasis. Current approaches are dependent on the subjective analysis of the radiologist measuring lymph node dimensions. This technology describes a three-fold approach to automatically identify lymph nodes in chest, abdomen and pelvis and to monitor them over time: 1) Detection of Lymph Nodes on Baseline CT Images, 2) Segmentation of Lymph Nodes on CT Images, 3) Matching of Lymph Nodes between two longitudinal CT images and segmentation of lymph nodes on one or more follow-up images.
Brain tumors are irregular, asymmetric and infiltrative. Standard measures of treatment response using diameter(s) are too simplistic, as information on tumor composition, its effects on surrounding tissue, and volume of tumor burden are not utilized. This invention builds on the previous technology (IR 2906) by adding new methods to automatically detect necrosis boundary inside a segmented brain tumor.
Prevention of breast cancer, the most common malignancy among women in the U.S., is a major public health issue. Breast density may serve as a useful intermediate biomarker for breast cancer risk assessment in investigations of potential chemopreventive agents for this disease. This invention provides an efficient tool to quantify changes in breast and fabric granular volumes on longitudinal MRI. This technology is fast and less affected by image noise. The software is semi-automatic and requires an operator to first loosely identify a breast region-of-interest. It can simultaneously segment several objects (i.e., the two breasts, multiple lesions and Fibroglandular tissues in breasts) on volumetric MRI and can analyze both axial and sagittal view breast images as well.
Lung cancer is the leading cause of cancer-related deaths in America. Early detection and treatment are key to reducing the mortality rate of lung cancer. Computer-aided detection of lung nodules on non-contrast-enhanced CT is helpful during the screening process. The present invention has an advantage over existing detection algorithms, especially in dealing with small ground glass opacity lung nodules and nodules attached to surrounding structures of similar density. Furthermore, the present invention is robust for a wide variety of CT imaging protocols used in both routine diagnosis and screening studies. This method has been tested on 294 CT scans in The Lung Image Database Consortium (LIDC) dataset. The sensitivity and false positive per scan for the training set (196/294) is 87% and 2.79. The sensitivity and false positive per scan for the testing set (98/294) is 85.2% and 3.22.
Brain tumors are irregular, asymmetric and infiltrative. Standard measures of treatment response using diameter(s) are too simplistic, as information on tumor composition, its effects on surrounding tissue, and volume of tumor burden are not utilized. This invention builds on the previous technology (IR 2906) by adding new methods to automatically detect necrosis boundary inside a segmented brain tumor.
Prevention of breast cancer, the most common malignancy among women in the U.S., is a major public health issue. Breast density may serve as a useful intermediate biomarker for breast cancer risk assessment in investigations of potential chemopreventive agents for this disease. This invention provides an efficient tool to quantify changes in breast and fabric granular volumes on longitudinal MRI. This technology is fast and less affected by image noise. The software is semi-automatic and requires an operator to first loosely identify a breast region-of-interest. It can simultaneously segment several objects (i.e., the two breasts, multiple lesions and Fibroglandular tissues in breasts) on volumetric MRI and can analyze both axial and sagittal view breast images as well.
Lung cancer is the leading cause of cancer-related deaths in America. Early detection and treatment are key to reducing the mortality rate of lung cancer. Computer-aided detection of lung nodules on non-contrast-enhanced CT is helpful during the screening process. The present invention has an advantage over existing detection algorithms, especially in dealing with small ground glass opacity lung nodules and nodules attached to surrounding structures of similar density. Furthermore, the present invention is robust for a wide variety of CT imaging protocols used in both routine diagnosis and screening studies. This method has been tested on 294 CT scans in The Lung Image Database Consortium (LIDC) dataset. The sensitivity and false positive per scan for the training set (196/294) is 87% and 2.79. The sensitivity and false positive per scan for the testing set (98/294) is 85.2% and 3.22.
Prompt assessment of stroke can help recover neurologic function that may have been lost during the acute phase. Infarct core volumes and infarct percentage may be a useful predictor for clinical outcome in acute stroke, but manual segmentation techniques limit its routine use. This technology describes a method and apparatus for accurately delineating and quantifying infarct area in two dimensions and volumes in three dimensions in medical images, including computed tomography (CT) and magnetic resonance images (MRI). It also provides an efficient mechanism for modifying segmentation results with minimal interaction. It can be integrated into routine, clinical workflow and thus can help predict clinic outcome and make non-invasive diagnoses.