Chapter title |
A New 2.5D Representation for Lymph Node Detection Using Random Sets of Deep Convolutional Neural Network Observations
|
---|---|
Chapter number | 65 |
Book title |
Medical Image Computing and Computer-Assisted Intervention – MICCAI 2014
|
Published in |
Lecture notes in computer science, September 2014
|
DOI | 10.1007/978-3-319-10404-1_65 |
Pubmed ID | |
Book ISBNs |
978-3-31-910403-4, 978-3-31-910404-1
|
Authors |
Holger R. Roth, Le Lu, Ari Seff, Kevin M. Cherry, Joanne Hoffman, Shijun Wang, Jiamin Liu, Evrim Turkbey, Ronald M. Summers, Roth, Holger R., Lu, Le, Seff, Ari, Cherry, Kevin M., Hoffman, Joanne, Wang, Shijun, Liu, Jiamin, Turkbey, Evrim, Summers, Ronald M. |
Abstract |
Automated Lymph Node (LN) detection is an important clinical diagnostic task but very challenging due to the low contrast of surrounding structures in Computed Tomography (CT) and to their varying sizes, poses, shapes and sparsely distributed locations. State-of-the-art studies show the performance range of 52.9% sensitivity at 3.1 false-positives per volume (FP/vol.), or 60.9% at 6.1 FP/vol. for mediastinal LN, by one-shot boosting on 3D HAAR features. In this paper, we first operate a preliminary candidate generation stage, towards -100% sensitivity at the cost of high FP levels (-40 per patient), to harvest volumes of interest (VOI). Our 2.5D approach consequently decomposes any 3D VOI by resampling 2D reformatted orthogonal views N times, via scale, random translations, and rotations with respect to the VOI centroid coordinates. These random views are then used to train a deep Convolutional Neural Network (CNN) classifier. In testing, the CNN is employed to assign LN probabilities for all N random views that can be simply averaged (as a set) to compute the final classification probability per VOI. We validate the approach on two datasets: 90 CT volumes with 388 mediastinal LNs and 86 patients with 595 abdominal LNs. We achieve sensitivities of 70%/83% at 3 FP/vol. and 84%/90% at 6 FP/vol. in mediastinum and abdomen respectively, which drastically improves over the previous state-of-the-art work. |
Twitter Demographics
Geographical breakdown
Country | Count | As % |
---|---|---|
United States | 4 | 33% |
India | 2 | 17% |
France | 1 | 8% |
Unknown | 5 | 42% |
Demographic breakdown
Type | Count | As % |
---|---|---|
Members of the public | 9 | 75% |
Scientists | 2 | 17% |
Practitioners (doctors, other healthcare professionals) | 1 | 8% |
Mendeley readers
Geographical breakdown
Country | Count | As % |
---|---|---|
United States | 2 | <1% |
Norway | 1 | <1% |
India | 1 | <1% |
Colombia | 1 | <1% |
Canada | 1 | <1% |
United Kingdom | 1 | <1% |
Japan | 1 | <1% |
China | 1 | <1% |
Unknown | 347 | 97% |
Demographic breakdown
Readers by professional status | Count | As % |
---|---|---|
Student > Ph. D. Student | 81 | 23% |
Researcher | 62 | 17% |
Student > Master | 61 | 17% |
Student > Bachelor | 24 | 7% |
Other | 16 | 4% |
Other | 51 | 14% |
Unknown | 61 | 17% |
Readers by discipline | Count | As % |
---|---|---|
Computer Science | 132 | 37% |
Engineering | 67 | 19% |
Medicine and Dentistry | 30 | 8% |
Agricultural and Biological Sciences | 9 | 3% |
Physics and Astronomy | 8 | 2% |
Other | 27 | 8% |
Unknown | 83 | 23% |