Confidence map generation for segmented optical coherence tomographic data

A non-transitory computer-readable storage medium comprising computer program instructions which, when executed by at least one processor, cause the at least one processor to perform a method of generating a segmentation confidence map by processing retinal layer segmentation data generated by a retinal layer segmentation algorithm, which generates the retinal layer segmentation data by calculating, for each voxel of at least a portion of a C-scan of a retina, a respective value of a classification indicator indicating a classification of the voxel as belonging to a retinal layer class of the predefined set of retinal layer classes, the method comprising:

• generating, for each voxel of a set of voxels for which the retinal layer segmentation data has been generated, a respective value of a first confidence indicator which is indicative of a level of confidence in the classification of the voxel;
• identifying, for a retinal layer class of the predefined set of retinal layer classes, a subset of the set of voxels such that the value of the classification indicator generated for each voxel of the subset indicates a classification of the voxel as belonging to the retinal layer class;
• calculating, for each A-scan of a plurality of A-scans of the C-scan, which A-scan has at least one voxel in the identified subset, a respective value of a second confidence indicator which is indicative of a level of confidence in a classification of the at least one voxel in the A-scan into the retinal layer class, based on at least one value of the first confidence indicator that has been respectively generated for the at least one voxel in the A-scan;
• and
• generating the segmentation confidence map using the calculated values of the second confidence indicator, such that the segmentation confidence map indicates a spatial distribution of a level of confidence in the classification of the voxels in the subset as belonging to the retinal layer class of the predefined set of retinal layer classes.

An apparatus for generating a segmentation confidence map by processing retinal layer segmentation data generated by a retinal layer segmentation algorithm, which generates, as the retinal layer segmentation data, a respective set of probability values for each voxel of at least a portion of a C-scan of a retina, wherein each probability value indicates a probability of the voxel belonging to a respective retinal layer class of a predefined set of retinal layer classes, the apparatus comprising:

• a voxel classification module arranged to generate, for each voxel of a set of voxels for which the retinal layer segmentation data has been generated:
• a respective value of a classification indicator based on the respective set of probability values, the value of the classification indicator indicating a classification of the voxel as belonging to a respective retinal layer class of the predefined set of retinal layer classes;
• and
• a respective value of a first confidence indicator which is indicative of a respective level of confidence in the classification of the voxel;
• a voxel identification module arranged to identify, for a retinal layer class of the predefined set of retinal layer classes, a subset of the set of voxels such that the value of the classification indicator generated for each voxel of the subset indicates a classification of the voxel as belonging to the retinal layer class;
• a confidence evaluation module arranged to calculate, for each A-scan of a plurality of A-scans of the C-scan, which A-scan has at least one voxel in the identified subset, a respective value of a second confidence indicator which is indicative of a level of confidence in a classification of the at least one voxel in the A-scan into the retinal layer class, based on at least one value of the first confidence indicator that has been respectively generated for the at least one voxel in the A-scan;
• and
• a segmentation confidence map generation module arranged to generate the segmentation confidence map using the calculated values of the second confidence indicator, such that the segmentation confidence map indicates a spatial distribution of a level of confidence in the classification of the voxels in the subset as belonging to the retinal layer class of the predefined set of retinal layer classes.

An apparatus for generating a segmentation confidence map by processing retinal layer segmentation data generated by a retinal layer segmentation algorithm, which generates the retinal layer segmentation data by generating, for each voxel of at least a portion of a C-scan of a retina, a respective value of a classification indicator indicating a classification of the voxel as belonging to a retinal layer class of the predefined set of retinal layer classes, the apparatus comprising:

• a confidence indicator evaluation module arranged to generate, for each voxel of a set of voxels for which the retinal layer segmentation data has been generated, a respective value of a first confidence indicator which is indicative of a level of confidence in the classification of the voxel;
• a voxel identification module arranged to identify, for a retinal layer class of the predefined set of retinal layer classes, a subset of the set of voxels such that the value of the classification indicator generated for each voxel of the subset indicates a classification of the voxel as belonging to the retinal layer class;
• a confidence evaluation module arranged to calculate, for each A-scan of a plurality of A-scans of the C-scan, which A-scan has at least one voxel in the identified subset, a respective value of a second confidence indicator which is indicative of a level of confidence in a classification of the at least one voxel in the A-scan into the retinal layer class, based on at least one value of the first confidence indicator that has been respectively generated for the at least one voxel in the A-scan;
• and
• a segmentation confidence map generation module arranged to generate the segmentation confidence map using the calculated values of the second confidence indicator, such that the segmentation confidence map indicates a spatial distribution of a level of confidence in the classification of the voxels in the subset as belonging to the retinal layer class of the predefined set of retinal layer classes.

A method of generating a segmentation confidence map by processing retinal layer segmentation data generated by a retinal layer segmentation algorithm, which generates, as the retinal layer segmentation data, a respective set of probability values for each voxel of at least a portion of a C-scan of a retina, wherein each probability value indicates a probability of the voxel belonging to a respective retinal layer class of a predefined set of retinal layer classes, the method comprising:

• generating, for each voxel of a set of voxels for which the retinal layer segmentation data has been generated:
• a respective value of a classification indicator based on the respective set of probability values, the value of the classification indicator (Iclass) indicating a classification of the voxel as belonging to a respective retinal layer class of the predefined set of retinal layer classes;
• and
• a respective value of a first confidence indicator which is indicative of a respective level of confidence in the classification of the voxel;
• identifying, for a retinal layer class of the predefined set of retinal layer classes, a subset of the set of voxels such that the value of the classification indicator generated for each voxel of the subset indicates a classification of the voxel as belonging to the retinal layer class;
• calculating, for each A-scan of a plurality of A-scans of the C-scan, which A-scan has at least one voxel in the identified subset, a respective value of a second confidence indicator which is indicative of a level of confidence in a classification of the at least one voxel in the A-scan into the retinal layer class, based on at least one value of the first confidence indicator that has been respectively generated for the at least one voxel in the A-scan;
• and
• generating the segmentation confidence map using the calculated values of the second confidence indicator, such that the segmentation confidence map indicates a spatial distribution of a level of confidence in the classification of the voxels in the subset as belonging to the retinal layer class of the predefined set of retinal layer classes.

A method of generating a segmentation confidence map by processing retinal layer segmentation data generated by a retinal layer segmentation algorithm, which generates the retinal layer segmentation data by calculating, for each voxel of at least a portion of a C-scan of a retina, a respective value of a classification indicator indicating a classification of the voxel as belonging to a retinal layer class of the predefined set of retinal layer classes, the method comprising:

• generating, for each voxel of a set of voxels for which the retinal layer segmentation data has been generated, a respective value of a first confidence indicator which is indicative of a level of confidence in the classification of the voxel;
• identifying, for a retinal layer class of the predefined set of retinal layer classes, a subset of the set of voxels such that the value of the classification indicator generated for each voxel of the subset indicates a classification of the voxel as belonging to the retinal layer class;
• calculating, for each A-scan of a plurality of A-scans of the C-scan, which A-scan has at least one voxel in the identified subset, a respective value of a second confidence indicator which is indicative of a level of confidence in a classification of the at least one voxel in the A-scan into the retinal layer class, based on at least one value of the first confidence indicator that has been respectively generated for the at least one voxel in the A-scan;
• and
• generating the segmentation confidence map using the calculated values of the second confidence indicator, such that the segmentation confidence map indicates a spatial distribution of a level of confidence in the classification of the voxels in the subset as belonging to the retinal layer class of the predefined set of retinal layer classes.