University of Southern California
Computer-aided Diagnosis (CAD)
Bone Age Assessment
Data Grid
RT Informatics
Informatics Beyond Radiology

Computer Aided Detection of Normal pressure hydrocephalus (NPH)


Classification and measurements of normal and abnormal intracranial compartments are the goals of segmentation procedures applied to quantitative evaluation of brain MR studies with specific regard to chronic degenerative disorders requiring serial controls over long period of time. Alzheimer disease (AD), normal pressure hydrocephalus (NPH), and multiple sclerosis (MS) serve as examples. NPH (fig.1) is a chronic accumulation of excess cerebrospinal fluid (CSF) within the ventricles (four inner, fluid-filled chambers) of the brain which causes the ventricles to dilate, stretching parenchymal (brain) tissue and resulting in a variety of clinical symptoms. NPH typically affects adults aged over 55, causing the classic triad of ataxia, urinary incontinence, and dementia. The symptoms of NPH appear to progress over time with a variable but often inevitable deterioration to disability or a critical loss of function. Although the exact incidence of NPH disorder in the American population is unknown, it is estimated that 375,000 people or 5% of the US population with dementia (including Alzheimer disease) may have NPH. Currently there is lack of accepted and standardized criteria for the diagnosis of NPH. Current diagnosis is based on convergent evidence from clinical history, examination and brain imaging. Brain imaging involves CT, MR or radionuclide cisternography to affirm ventriculomegaly and/or abnormal CSF flow. Three soft qualifiers: probable, possible, and unlikely are utilized in the current classification scheme of NPH. Shunt placement (surgical decompression) is the treatment of choice for the select group patients determined to likely benefit from invasive therapy. Initial evaluation of surgical candidacy is very important. In 70-75% of cases of qualifying patients subjected to shunt surgery, there is significant clinical improvement following intervention, while in the remaining cases there is no improvement. Selection of shunt candidacy is presently supported by patient history and 2D/3D manual measurements in MR/CT studies. Among these measurements, the volume of the lateral ventricles plays an important role. This finding is also used in the assessment of shunt patency for management of patients who failed to improve or are clinically deteriorating. Routine clinical protocol for brain imaging of NPH patients sometimes includes cross-sections thicker than 3mm with gaps between cross-sections.


An automated approach to lateral ventricles segmentation from T1/T2 head MR studies is a goal of our research. The methodology includes head mask and interhemispheric lines location followed by a 3D fuzzy c-means (FCM) segmentation of a 3D lateral ventricles region from T1 images. Segmented CSF spaces are subjected to a morphological analysis preserving connected components in 3D space and yielding lateral ventricles LV volume. This approach is developed and tested on T1/T2 brain MRI studies of 40 control cases and 7 NPH cases with 5mm cross sections and 1.25 mm gaps.

Figure 1. Examples of ventricular system in MR imaging: a), b) T1/T2 images of dilated ventricular system of 68 year old male NPH patient, c), d) T1/T2 images of ventricular system within normal limits of 54 year old female.


A movie with segmented lateral ventricles of an NPH patient.

A movie with segmented lateral ventricles of a non-NPH patient.

Copyright 2005-2013.  Image Processing and Informatics Lab, 1042 Downey Way. Denney Research Center (DRB) 264 Los Angeles, CA 90089-1111   Phone: 213.821.8395 Fax: 213.743.2962