Misdiagnosis of community-acquired pneumonia is an important clinical problem, leading to a high rate of mortality. Diag- noses are typically conducted using two-dimensional chest x-rays, which have shown to be time-consuming and inaccurate. In an effort to improve the current diagnostic method, we utilized Micro-Computed Tomography (MicroCT) and image analysis software to develop a diagnostic algorithm that can quantitatively assess the severity of pneumonia in mice. We believed this method would provide more immediate, precise, and accurate diagnoses as opposed to the qualitative assessments done by ra- diologists at present, because MicroCT provides opportunities for non-invasive radiographic endpoints for pneumonia studies. A quantitative scoring of previously obtained Computed Tomography (CT) scans of pneumonia infected and control mice lungs was developed with a semi-automated image segmentation algorithm. At the endpoint of 168 hours, each of the mice was categorized as either a) a Saline (control)-injected mouse (total=13), a Pneumonia-injected Survivor (total=11), or a Pneumo- nia-injected Non-survivor (total=11). Three comparison tests were then completed, including Saline vs. All Pneumonia Injected Mice, Pneumonia Survivors vs. Pneumonia Non-survivors, and All Survivors (both Saline & Pneumonia) vs. Pneumonia Non-survivors. In all three comparisons, the semi-automated algorithm was better able to distinguish between the different groups than radiologists using two-dimensional chest x-rays of the mice’s lungs, with p-values of 0.001, 0.039, and 0.001 for the semi-automated algorithm, and 0.004, 0.581, 0.058 for the radiologists, respectively. Key Words: Community-acquired pneumonia, Computed Tomography Columbia Undergraduate Science Journal Open-Access Publication | http://cusj.columbia.com 4 cusjVolume 5Spring 2011 Columbia Undergraduate Sci J http://cusj.columbia.edu B IO M ED IC A L EN G IN EE R IN G cusjcolumbia undergraduate science journalResearch Articles 5cusj Volume 5Spring 2011 Columbia Undergraduate Sci J http://cusj.columbia.edu B IO M ED IC A L EN G IN EER IN G In the United States, pneumonia is the sixth lead- ing cause of death and the number one infectious disease killer (M. S. Niederman, 1998). !e disease is an in"ammation of one or both lungs caused by an infection from bacteria, viruses or fungi. !e infection causes the alveoli of the lungs to become in"amed and #lled with "uid, which leads to symptoms such as cough, fever and respiratory breathing di$culties (Jelic, 2005). Often times, pneumonia occurs as a secondary infection when the immune system of a person is already weakened due to prior infection, such as an upper respiratory tract infection. !is primary infection causes in"ammation in the inner lining of airways that leaves the patient susceptible to the secondary infections such as pneumonia (Boone, 2004). Pneumonia can be classi#ed according to the population a%ected. Hospital-acquired pneumonia is acquired when a patient breathes germs during a hospital stay for another illness. People are most prone to hospital-acquired pneumo- nia while on a mechanical ventilator, since potentially pneu- monia-causing bacteria and viruses may be blown directly into the lungs. !e most common type of pneumonia is community-acquired. Community-acquired pneumonia is acquired outside of hospitals and other health care settings, with about 5.6 million people getting infected every year in the USA and 1.1 million requiring hospitalization (M. S. Niederman, 1998). Community-acquired pneumonia is an important clinical problem, with high rates of misdiagnosis and mortality. Current methods to diagnose pneumonia rely on two dimensional (2D) chest X-rays, which are known to have low sensitivity early in the course of pneumonia (Mohd). Radiologists typically score six lung zones (upper, middle, and lower, on the right and left sides) for each pa- tient on a scale of 0 to 4, such that zero is normal, and the maximum possible abnormal score is 24 for the combined zones; 0 represents 0% pneumonia involvement, 1 repre- sents up to 25% involvement, 2 represents up to 50%, 3 represents up to 75%, and 4 represents up to 100% (Arm- brust, 2005). !ese chest X-rays may take days to diagnose the severity of pneumonia, in which time immunocompro- mised patients, such as patients with HIV/AIDS, cancer, diabetes, or sickle cell anemia, may reach a severity beyond curing (Smergal, 2008; Stuart, 2008). For example, immu- CT Based Semi-Automated Method for Pneumonia Severity in Mice Ansh Johri1*, Lewis Hsu 2 1Deparment of Biomedical Engineering, Columbia University, New York, NY 10027; 2National Institutes of Health, Nation- al Heart, Lung, Blood Institute, Washington, DC 20010. Copyright: © 2011 The Trustees of Columbia University, Co- lumbia University Libraries, some rights reserved, Johri, et al. Received Dec. 31, 2009. Accepted Feb. 8, 2010. Published April 1, 2011. *To whom correspondence should be addressed: Center for Cancer and Blood Disorders, West 4-600. Children’s National Medical Center111 Michigan Ave., N.W. Washing- ton, DC 20010 aj2381@columbia.edu Abstract Introduction nocompromised patients with pneumonia have a mortality rate of 12% (Mohd). Furthermore, radiologists are often inconsistent with their diagnoses; two radiologists may judge the severity of pneumonia in patients very di%erently, leading to possible misdiagnosis (L Hsu, 2007). !us, im- aging techniques to evaluate pneumonia earlier and with more accuracy would be important diagnostic tools for cli- nicians. Imaging information could also be used to guide decisions on the clinical care needed, such as whether to hospitalize or to treat the patient at home, thus improving pneumonia diagnosis. In order to address these limitations of inaccuracy, in- consistency, and delayed diagnosis, a di%erent diagnostic method is required. Computed Tomography (CT) scans use X-rays that pass through the specimen and are received by sensors on the other end. Denser portions of the speci- men result in a reduced amount of radiation received by the other end, since the specimen hinders the radiation. !is disparity in densities, or attenuation, can be reconstruct- ed to produce a 3D image with di%erent grayscale values (Figure 1). Houns#eld Units (HU) are grayscale values that correspond to the density of each voxel. In the Houns#eld scale, -1000 represents air, 0 represents water, and 1000 represents bone density. Notably, "uid or pus in the alveo- lar sacs would be approximately 0 HU, normal lung alveoli have a mixture of air and tissue reading near -500 HU, and voxels in lung with a mixture of air and "uid would be between -500 and 0 HU. CT scans, which can visualize the entire lung as opposed to the 2D projection scans in a chest radiograph, might have the sensitivity to assess the severity of pneumonia as early as 24 hours after onset. !is earlier timeframe for treatment would allow immunocom- promised patients to receive immediate treatment, thus de- creasing their mortality rate. Since CT scans provide a more detailed depiction of the lung, they are potentially more accurate than the current chest X-ray method. Finally, by developing a semi-automated method that uses CT scans to diagnose the severity of pneumonia, more precise diagnoses can be conducted, since the procedure is more automated and less prone to human error (Muller, 2006). !e purpose of this particular research was to develop an algorithm to measure the severity of pneumonia in mice through Micro-Computer Tomography (MicroCT) Scan Analysis and test its e%ectiveness through comparison with radiologists’ diagnosis. MicroCT works in the same way as a regular CT scanner, but is typically used to image smaller specimens, such as rodents, as opposed to human beings. !ere were three goals for the image analysis algorithm. !e #rst was to achieve high reproducibility in repeated analysis of the same MicroCT scan. Current methods typi- cally involve having two radiologists independently score the chest X-rays; the #nal score is then the average of the independent scores. !e second goal was to achieve higher accuracy using image segmentation algorithm to quantify the amount of pneumonia in the lungs. !is is di%er- ent from current methods which require radiologists to qualitatively assess multiple images of pneumonia. !e quanti#cation would be done by loading the CT scans in an imaging software, and determining the voxel distribution in order to compare densities. Fi- nally, the project aimed to increase e$ciency in diag- nosis. A semi-automated computer algorithm would allow more measurements to be taken in a smaller amount of time than with current methods, without special expertise in radiology. A computer would au- tomatically calculate the severity of the pneumonia, which, under current circumstances, would be done by a radiologist. In order to use the computer algo- rithm method, it would be necessary for the radiologist to have some basic skills, however. !e #rst is the ability to use Amira, the software used in this paper. !e second is knowledge of basic lung anatomy, such as the location of the trachea, stomach bubble, and mediastinum. Finally, the radiologist would need the ability to use a quantitative diag- nosis performed by the computer to give the correct treat- ment to the patient. We hypothesized that in vivo MicroCT scans of mice with early bacterial pneumonia could be scored quantita- tively by semi-automated imaging methods, with good re- producibility and correlation with the bacterial dose inocu- lated, pneumonia survival outcome, and radiologists’ scores previously obtained. Materials and Methods !e project used MicroCT scans to evaluate a murine model of bacterial pneumonia through image analysis by semi-automated segmentation and comparison of results to Figure 1 �+�9LJVUZ[Y\J[LK�3\UN��