Presentation is loading. Please wait.

Presentation is loading. Please wait.

Variability of LIDC Panel Segmentations and Soft Segmentation of Lung Nodules Presented by Stephen Siena and Olga Zinoveva.

Similar presentations


Presentation on theme: "Variability of LIDC Panel Segmentations and Soft Segmentation of Lung Nodules Presented by Stephen Siena and Olga Zinoveva."— Presentation transcript:

1 Variability of LIDC Panel Segmentations and Soft Segmentation of Lung Nodules Presented by Stephen Siena and Olga Zinoveva

2 Overview Discussion of LIDC data New variability metric Soft segmentation ◦ Related work ◦ Methods ◦ Discussion Future Work

3 The LIDC Chest CT scans reviewed by 4 radiologists ◦ Semantic characteristics and contours Benefits to research ◦ Access for everyone ◦ Sets standard Problem: No ground truth ◦ No perfect detection/outline of nodules

4 Our Proposed Solution Find “accuracy” of radiologists first Provides measure of panel segmentations’ consistency ◦ Validation of reference truth Incorporates and improves upon previous metrics

5 Methodology Cost matrix Variability matrix

6 Methodology Variability index Normalized variability index

7 (a) Radiologist outlines (b) Pmap (c) Cost matrix (R = 4; k = 10) (d-g) Variability matrix after 0, 1, 2, 3 iterations (h)Final variability matrix VI = 60; VI n = 5.1064

8 Results VI n =.5227VI n = 1.8198 VI n = 15.1705VI n = 37.8774 VI n = 3.2339 VI n = 7.0345

9 Complements Overlap Overlap =.2245 VI n = 11.6000 Overlap =.2246 VI n = 35.4449 Overlap =.2064 VI n = 81.4449 Overlap =.2763 VI n = 12.3711 Overlap =.4462 VI n = 12.5101 Overlap =.6771 VI n = 12.5896

10 Background Most lung nodule segmentation algorithms produce hard segmentations Probabilistic segmentations used for other medical imaging purposes ◦ Cai, Hongmin et al. Produced brain segmentations ◦ Tang, Hui – kidney segmentations van Ginneken, Bram produced lung nodule segmentations on the first LIDC dataset

11 Dataset and pre-processing All slices from the LIDC 85 dataset that contain four radiologist contours ◦ 264 slices representing 39 nodules Different CT scanners convert HU to intensity differently ◦ Solution – intensity shift based on the rescale intercept

12 Random point selection 0%, 25%, 50%, 75%, 100% Points selected proportionately from every region and every image.

13 Random point selection Coordinates selected randomly, but must be at least R pixels away from each other for any region P inT is the total number of pixels in agreement area i of image n P ins is the number of pixels selected from agreement area i of image n

14 Classifier Intensity and texture (Gabor and Markov) features calculated for a 9X9 neighborhood around each pixel Classifier assigned a continuous probability (0-100) for each pixel’s membership in the nodule class These values were thresholded to produce a p-map of the segmentation

15 Classifier results Median soft overlap: 0.53 Median VI: 4.24 (Q1:2.5, Q2:9.8) Chest wall causes the majority of errors Over-segmentation on most slices

16 Classifier results NoduleRadiologist p-mapClassifier’s p-map

17 Post-processing: VI Trimming Check if any of the values are above the threshold, reset them to 0 in the p-map and update the pointer matrix Pointer matrix still contains positive values All values in pointer matrix are non-positive Expand the existing point or matrix with output from the classifier and calculate the variability matrix for this p-map Begin with a seed point Terminate

18 VI Trimming: Example 033 043 012 1011 01 32 12321 2 000 2 3 000 3 2 000 2 12321 01112 00332 00432 00122 00001 144443 10113 2010013 2010323 201312 5 1233321 2 00000 2 2 00000 3 1 00000 3 0 00000 3 0 00000 2 0123321

19 Results after VI trimming Median soft overlap: 0.57 (vs. 0.53) Median RAE: 9.9 (vs. 17.5)

20 Results after VI trimming

21 Ongoing and future work Improve segmentation of nodules attached to chest walls Select seed points without manual input Calculate VI for the segmentations Work with all slices for the 39 nodules Expand to the new LIDC dataset Segment lungs ◦ Eliminates the need for chest wall separation algorithms ◦ Allows for better intensity normalization

22 References Cai, Hongmin et al. “Probabilistic Segmentation of Brain Tumors Based on Multi-Modality Magnetic Resonance Images,” 4th IEEE International Symposium on Biomedical Imaging: From Nano to Macro 600-603 (April 2007) Tang, Hui et al. “A vectorial image soft segmentation method based on neighborhood weighted Gaussian mixture model,” Computerized Medical Imaging and Graphics (August 2009) van Ginneken, Bram, “Supervised Probabilistic Segmentation of Pulmonary Nodules in CT Scans,” MICCAI 2006 Proceedings, Part II pp. 912-919 (October 2006)

23 Questions?


Download ppt "Variability of LIDC Panel Segmentations and Soft Segmentation of Lung Nodules Presented by Stephen Siena and Olga Zinoveva."

Similar presentations


Ads by Google