Introduction Anotation Segmentation Detection 2 Nodule interpretation characteristics Characteristic Possible Scores Calcification 1 Popcorn 2 Laminated 3 Solid 4 Noncentral 5 Central ID: 415027
Download Presentation The PPT/PDF document "Outline" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Outline
Introduction
Anotation
Segmentation
DetectionSlide2
2
Nodule interpretation (characteristics)
Characteristic
Possible Scores
Calcification1. Popcorn2. Laminated3. Solid4. Non-central5. Central6. AbsentInternal structure1. Soft Tissue2. Fluid3. Fat4. AirLobulation1. Marked2. . 3. . 4. .5. NoneMalignancy1. Highly Unlikely 2. Moderately Unlikely3. Indeterminate4. Moderately Suspicious 5. Highly Suspicious
CharacteristicPossible ScoresMargin1. Poorly Defined2. . 3. . 4. .5. SharpSphericity1. Linear2. .3. Ovoid4. .5. RoundSpiculation1. Marked2. . 3. . 4. .5. NoneSubtlety1. Extremely Subtle 2. Moderately Subtle 3. Fairly Subtle 4. Moderately Obvious 5. ObviousTexture1. Non-Solid2. .3. Part Solid/(Mixed) 4. .5. Solid
7 out of 9 semantic characteristics have a broad range of values for the 149 nodulesSlide3
Interpretation
Not only ratings, but also boundaries are different
Reader 1
Reader 2Reader 3Reader 4Lobulation - 4Malignancy - 5 Margin - 4Sphericity - 2Spiculation - 1Subtlety - 5Texture - 4Lobulation - 1Malignancy - 5Margin - 3Sphericity - 4Spiculation - 2Subtlety - 5Texture - 5Lobulation - 2Malignancy - 5Margin - 3Sphericity - 5Spiculation - 2Subtlety - 5Texture - 4Lobulation - 5Malignancy - 5Margin - 2Sphericity - 3Spiculation - 4Subtlety - 5Texture - 4Slide4
Proposed methodology
The automatic mapping extraction is:
SEMI-SUPERVISED
Only small amount of data is initially labeled. Based on ACTIVE LEARNINGIteratively adds data to the training set. Slide5
5
Methodology: Ensemble of classifiers (Active-Decorate)Slide6
6
Results (Accuracy)
Characteristics
Decision trees
Add instances predicted with high confidence (60%)Add instances predicted with high confidence (60%) and instances with low margin (5%)Lobulation27.44%81.00%69.66%Malignancy42.22%96.31%96.31%Margin35.36%98.68%96.83%Sphericity36.15%91.03%
90.24%Spiculation36.15%63.06%58.84%Subtlety38.79%93.14%92.88%Texture53.56%97.10%97.36%Average38.52%88.62%86.02%Slide7
Radiology reportSlide8
IRMA
- T (technical): image modality
- D (directional): body orientation - A (anatomical): body region examined
- B (biological): biological system examined This allows a short and unambiguous notation (IRMA: TTTT – DDD – AAA – BBB),Slide9
IRMASlide10
Questions?