SIFT Key-Points for Lung Tumor Detection in PET/CT Images

Nitschke M, Jian T, Bayer S, Ren Q, Maier A (2017)


Publication Type: Conference contribution, Abstract of a poster

Publication year: 2017

Event location: Edinburgh

URI: https://www5.informatik.uni-erlangen.de/Forschung/Publikationen/2017/Nitschke17-SKF-talk.pdf

Abstract

Automatic tumor segmentation enables a fast, reliable and reproducible diagnostic to deliver best treatment procedures for cancer patients. However, state-of-the-art tumor segmentation algorithms are based on manual input which is time- and work-consuming. Furthermore, varying inputs can lead to different segmentation results. To simplify the workflow, a tumor detection algorithm should be used instead of manual input. In this preliminary study, a tumor detection algorithm was investigated and validated for six PET/CT data sets of lung tumor patients. Scale-invariant feature transform (SIFT) keypoints were extracted as tumor candidates in PET images. For each candidate point, intensity, scale and orientation invariant features were calculated both from the PET images and registered CT images. Based on these features, the keypoints were classified either as tumor or as background with a neural network. The neural network classification showed an AUC of 0.71. The tumor detection ratio was 78.6 %. The result of the evaluation with clinical data demonstrated the high potential of the presented lung tumor detection algorithm. We are convinced, further developments of this approach will result in enhanced cancer detection in the entire body.

Authors with CRIS profile

Additional Organisation(s)

Involved external institutions

How to cite

APA:

Nitschke, M., Jian, T., Bayer, S., Ren, Q., & Maier, A. (2017, July). SIFT Key-Points for Lung Tumor Detection in PET/CT Images. Poster presentation at Medical Image Understanding and Analysis (MIUA), Edinburgh.

MLA:

Nitschke, Marlies, et al. "SIFT Key-Points for Lung Tumor Detection in PET/CT Images." Presented at Medical Image Understanding and Analysis (MIUA), Edinburgh 2017.

BibTeX: Download