Evaluating quality of segmentation of an image into different types of tissue for planning treatment using tumor treating fields (TTFields)
To plan tumor treating fields (TTFields) therapy, a model of a patient's head is often used to determine where to position the transducer arrays during treatment, and the accuracy of this model depends in large part on an accurate segmentation of MRI images. The quality of a segmentation can be improved by presenting the segmentation to a previously-trained machine learning system. The machine learning system generates a quality score for the segmentation. Revisions to the segmentation are accepted, and the machine learning system scores the revised segmentation. The quality scores are used to determine which segmentation provides better results, optionally by running simulations for models that correspond to each segmentation for a plurality of different transducer array layouts.
1. A method of determining a quality of a segmentation of an image of a head that includes a plurality of tissue types, the method comprising the steps of:
measuring a Dice coefficient between a computed head segmentation and a validated head segmentation of a training set;
estimating a registration quality between the computed head segmentation and the validated head segmentation;
using a shortest axis length and a signal to noise ratio of each of the tissue types separately as an indicator of input image quality; and
comparing the measured Dice coefficient to a plurality of predictions based on Dice coefficients'.
2. The method of claim 1 , wherein the registration quality is estimated based on a deformation field bias, directional variability, and mean per-axis variability.
3. The method of claim 1 , wherein a volume of a shape and a number of connected components are used to describe a segmentation shape.
4. The method of claim 1 , further comprising outputting a result of comparing the measured Dice coefficients to the plurality of predictions based on Dice coefficients as a quality score of the computed head segmentation.
5. The method of claim 1 , further comprising outputting a result of comparing the measured Dice coefficients between the sets to the plurality of predictions based on Dice coefficient as a quality score of the computed head segmentation for an entirety of the input image.
6. The method of claim 1 , further comprising outputting a separate result of comparing the measured Dice coefficients between the sets to the plurality of predictions based on Dice coefficients a quality score of the computed head segmentation for each of a plurality of regions within the input image.
7. The method of claim 1 , further comprising outputting a separate result of comparing the measured Dice coefficients between the sets to the plurality of predictions based on Dice coefficients as a quality score of the computed head segmentation for each of the tissue types.
8. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients are generated using features describing a segmentation shape.
9. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients' predictions are generated using features, each of which is computed for each of the tissue types, describing a segmentation shape.
10. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients' predictions are generated using a volume of a segmentation shape and a number of connected components to describe the segmentation shape.
11. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients' predictions are generated using a volume of a segmentation shape and a number of connected components, each of which is computed for each of the tissue types, to describe the segmentation shape.
12. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients' predictions are generated using a decision tree regressor.
13. The method of claim 1 , wherein the measuring of the Dice coefficient comprises measuring a separate Dice coefficients for each of the tissue types.
14. The method of claim 1 , wherein the plurality of predictions based on Dice coefficients' predictions are generated for each of the tissue types.
15. The method of claim 1 , wherein the measuring of the Dice coefficient is performed between a computed head segmentation and a validated head segmentation of a plurality of training sets.
16. The method of claim 1 , wherein a volume of a shape and a number of connected components, each of which is computed per tissue, are used to describe a segmentation shape.