Evaluation Of Semi-Automatic Region Of Interest Detection Method For Glomerular Filtration Rate Quantification

Glomerular Filtration Rate （GFR） is an important index in evaluating renalfunction. There existed several approaches for GFR determination including inulinclearance, blood plasma clearance, and gamma camera renography with radioactivetracer. Among these methods gamma camera renography is the most commonly usedone which estimated GFR via the Gates method after the delineation of kidney regionof interest （ROI）. In routine clinical practice, the kidney ROI is defined manually dueto the lack of an existing automated detection procedure. Apparently, manual ROIdelineation method is subjective and time-consuming, making the ROI acquisitionmore difficult and apt to error, especially for patients with low renal functions.Objectives: The purpose of this work is to introduce a novel ROI detectionmethod adapted to the quality and presence of certain features on the functional imagesdue to the severity of kidney damages. For kidneys with mild damage, ROIs on thecorresponding functional SPECT images alone can be identified relatively easily andsemi-automatically, Otherwise, for kidneys that are damaged moderately or severelyand with blurred contour on the functional images, additional anatomical informationfrom CT images is incorporated to assist the semi-automatic ROI detection.Methods: A composite image was formed with dynamic frames acquired duringthe2to3min time interval after tracer injection. For the semi-automatic ROI detectionmethod, image portions containing liver, heart, spleen and big vessels were firstremoved from the composite image based on the temporal profile of traceraccumulation in these organs. The image was then smoothed with a Gauss filter and amaximum filter, followed by a morphological processing step. For the CT-assistedsemi-automatic ROI detection method, separate kidney ROIs from composite imagesand CT images were first delineated manually. Subsequently, the locate-matching algorithm was employed to map the targeted ROI from CT image to the compositeimage. Finally, the kidney ROI and the background ROI were obtained, from whichGFR was determined based on the counts derived.Results: Data from forty-seven patients were randomly selected from a clinicaldatabase. All patients had damaged kidneys with split renal function （SRF） lower than50ml/min/1.73m². Using the SPECT based semi-automatic ROI detection method,ROIs were obtained successfully and the GFR based on this semi-automated approachcorrelate well with the one obtained via the standard （r1=0.957, r₂=0.951）. Additionalnineteen patients diagnosed as moderately or severely damaged kidney with SRF lowerthan30ml/min/1.73m²were also included to validate the effectiveness and feasibility ofthe proposed CT-assisted semi-automatic ROI detection method. The GFR computationshows that the CT-assisted method is reproducible and has good consistency betweendifferent operators.Conclusion: This work presented a semi-automatic ROI detection methodemploying different algorithms to deal with subjects with different severity of kidneydamages. The procedure is easy to use, objective and with the high accuracy andreproducibility of the estimated GFR values.