|
The efficacy of computer detection of kidney in PET/CT images
and kidney SUV statistics in PET images |
|
X Huang, PHD, |
|
PURPOSE |
|
To determine whether a computer-aided detection program can
detect internal organs such as the kidney in PET/CT images, and to evaluate
the efficacy of discounting normal physiological uptake of FDG by these
organs in PET images. |
|
|
|
METHOD AND MATERIALS |
|
Sixty-six patients with lung cancer, liver cancer, breast
cancer, or lymphoma were imaged using PET/CT. The ''hot spots'' in PET
correspond to either normal physiological uptake (such as by kidney, heart or
bladder), or abnormal pathological uptake (such as due to tumors). We develop
a computer program that automatically detects internal organs such as the
kidney, bladder, heart and liver. The program is designed based on a learning
framework in which we train a discriminative classifier that recognizes a
local region containing the organ of interest. In kidney detection, if the
detected region corresponds to a hot spot (SUV >2.5) in PET, the hot spot
is interpreted as relating to physiological uptake by the kidney. Then a
segmentation method is applied to delineate the boundary of the kidney hot
spot, and statistics on kidney hot-spot SUV values are collected. |
|
|
|
RESULTS |
|
Using forty-six patient data as training, and twenty patient data
as testing, the kidney detection rate is 90%. In the detected kidney regions,
the maximum SUV value is between 3.8~19.3, the average maximum SUV value is
12.5, and the variance in maximum SUV value is 6.2. The mean SUV value in
kidney regions is between 2.8~15.6, the average mean SUV value is 8.7 and the
variance in mean SUV value is 4.2. |
|
|
|
CONCLUSION |
|
The SUV statistics of kidney in PET images overlap with reported
statistics of carcinoma, lymphoma and other types of tumors. Computer systems
can be trained to detect internal organs that have high physiological FDG
uptake such as the kidney, hence enabling hot spots that correspond to
physiological uptake (such as in kidney) to be accurately interpreted and
discounted, leaving only hot spots that correspond to pathological uptake
(such as in tumors) in PET images. |
|
|
|
CLINICAL RELEVANCE/APPLICATION |
|
Computer-aided detection of internal organs such as kidney,
bladder, heart in PET/CT images and hot spot interpretation in PET images. |
|
|