Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||Automatic detection of lung tumor and abnormal regional lymph nodes in PET-CT images|
|Publisher:||Society of Nuclear Medicine|
|Source:||The Journal of nuclear medicine, 2011, v. 52, no. suppl. 1, 211 How to cite?|
|Journal:||The Journal of nuclear medicine|
|Abstract:||Objectives: To develop an automatic approach to detect primary lung tumor and disease in regional lymph nodes from FDG PET-CT images of the thorax.|
Methods: For each PET and CT image, a texture feature set was obtained using a filter bank of 24 Gabor filters. The mean-shift clustering algorithm was then applied in the feature space to cluster each image into regions. In the PET images, separation of regions highlighted the discontinuities in uptake levels. The CT image regions outlined the structural information. The regions in the PET images with average SUV higher than a defined threshold were then identified as tumor or abnormal lymph nodes. Their boundaries were further refined based on the contours of the spatially corresponding CT regions, if they overlapped with the identified PET regions by at least 90%. The threshold was defined as the sum of (1) the average SUV of the region with the highest uptake (weighted by 0.15 ), and (2) the average SUV of regions approximating the mediastinum.
Results: The method was tested on 40 NSCLC PET-CT studies and compared to three SUV threshold-based approaches: SUV-2.5, 50% SUVmax, and (0.15*SUVmean+SUVbkg) . A correct detection of tumor or abnormal lymph nodes with acceptable boundary delineation was considered as true positive. As shown in Figure 1, our method exhibited the highest accuracy. Its high precision is attributed to it particularly effectively filtering out the relatively high uptake and non-pathological regions in the mediastinum, while the suboptimal recall is due to false negative detections of small regions. Visual inspection showed that the boundaries approximated the structures depicted on CT more closely with our method.
Conclusions: The developed method demonstrates higher performance for detecting lung tumor and disease in regional lymph nodes from PET-CT images compared to the benchmarked approaches.
|Appears in Collections:||Journal/Magazine Article|
Show full item record
Checked on May 21, 2017
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.