Method of non-invasive detection of tumour and/or healthy tissue and hyperspectral imaging apparatus
A method of non-invasive detection of tumor and/or healthy tissue using hyperspectral imaging is provided. The method includes determining a pixel-wise classification of a hyperspectral image obtained from a target tissue by implementing a machine learning approach fed with hyperspectral data of at least one pixel. The method further includes performing a segmentation of the hyperspectral image by grouping pixels into at least one cluster based on the hyperspectral data of each pixel. The same classification is assigned to all the pixels belonging to the same cluster, so that each cluster is associated with a certain classification.
1. A method of non-invasive detection of tumor and/or healthy tissue using hyperspectral imaging, the method comprising:
determining a pixel-wise classification of a hyperspectral image obtained from a target tissue by implementing a machine learning approach fed with hyperspectral data of at least one pixel;
performing a segmentation of the hyperspectral image by grouping pixels into at least one cluster based on the hyperspectral data of each pixel;
assigning the same classification to all the pixels belonging to the same cluster, so that each cluster is associated with a certain classification; and
performing a pre-processing of the hyperspectral data, prior to at least one of determining the pixel-wise classification, performing the segmentation, or assigning the same classification.
2. The method according to claim 1 , wherein the method further comprises:
performing a spatial homogenization of the pixel-wise classification, prior to assigning the same classification for all the pixels belonging to the same cluster.
3. The method according to claim 2 , wherein the method further comprises:
generating a spatially homogenized pixel-wise classification map of the hyperspectral image, after performing a spatial homogenization of the pixel-wise classification.
4. The method according to claim 2 , wherein the spatial homogenization of the pixel-wise classification comprises:
performing a dimensionality reduction of the hyperspectral data of each pixel;
implementing a filtering approach fed with pixel-wise classification data and dimensionally-reduced hyperspectral data, the filtering approach assigning to a pixel the most common classification among its pixel-neighbours.
5. The method according to claim 1 , wherein the method further comprises:
generating a pixel-wise classification map of the hyperspectral image, after determining the pixel-wise classification.
6. The method according to claim 1 , wherein the method further comprises:
generating a segmentation map of the hyperspectral image, after performing the segmentation of the hyperspectral image.
7. The method according to claim 1 , wherein the method further comprises:
generating a cluster-wise classification map of the hyperspectral image, after assigning the same classification to all the pixels belonging to the same cluster.
8. The method according to claim 7 , wherein the method further comprises:
assigning in the cluster-wise classification map different chromatic properties to clusters with different classifications to each other, predefined chromatic properties being related to each classification.
9. The method according to claim 8 , wherein the method further comprises:
degrading chromatic properties of a cluster depending on the probability of the assigned classification.
10. The method according to claim 8 , wherein the method further comprises:
mixing different chromatic properties related to different classifications in the same cluster, taking into account the percentage of the probability of different classifications in the cluster.
11. The method according to claim 1 , wherein the classification comprises at least one of the following classes: tumor tissue, healthy tissue, hypervascularized tissue, blood vessel, background element, primary tumor tissue and metastatic tumor tissue.
12. The method according to claim 1 , wherein a classification is assigned to the cluster corresponding to the most frequent pixel classification in each cluster.
13. The method according to claim 1 , wherein the pre-processing of the hyperspectral data comprises at least one of the following: image calibration, noise filtering and noise removing, band averaging and normalization.
14. The method according to claim 13 , wherein the image calibration comprises correcting variations in the spectral signature of each pixel caused by non-uniform illumination over a surface of the target tissue.
15. The method according to claim 13 , wherein the band averaging comprises removing at least one band of the hyperspectral image, the removed band being correlated with another consecutive band.
16. The method according to claim 13 , wherein the normalization comprises homogenizing the amplitude of each pixel without modifying the shape of its spectral signature.
17. A hyperspectral imaging apparatus comprising:
an optical device configured to display a visible image of a target tissue;
a camera configured to obtain hyperspectral images of the target tissue;
a processing platform in data communication with the camera and the optical device;
the optical device being also configured to display information obtained from processed hyperspectral images;
the processing platform being configured to implement a method of non-invasive detection of tumor and/or healthy tissue using hyperspectral imaging, the method comprising:
determining a pixel-wise classification of a hyperspectral image obtained from a target tissue by implementing a machine learning approach fed with hyperspectral data of at least one pixel;
performing a segmentation of the hyperspectral image by grouping pixels into at least one cluster based on the hyperspectral data of each pixel;
assigning the same classification to all the pixels belonging to the same cluster, so that each cluster is associated with a certain classification; and
performing a pre-processing of the hyperspectral data, prior to at least one of determining the pixel-wise classification, performing the segmentation, or assigning the same classification.
18. The hyperspectral imaging apparatus according to claim 17 , wherein the optical device comprises a surgical microscope; the surgical microscope comprising one eyepiece configured to display the visible image of the target tissue and another eyepiece configured to display the information obtained from processed hyperspectral images.
19. A computing device comprising a memory and a processor, embodying instructions stored in the memory and executable by the processor, the instructions comprising functionality to execute a method of non-invasive detection of tumor and/or healthy tissue using hyperspectral imaging of non-invasive detection of tumor and/or healthy tissue using hyperspectral imaging, the method comprising:
determining a pixel-wise classification of a hyperspectral image obtained from a target tissue by implementing a machine learning approach fed with hyperspectral data of at least one pixel;
performing a segmentation of the hyperspectral image by grouping pixels into at least one cluster based on the hyperspectral data of each pixel;
assigning the same classification to all the pixels belonging to the same cluster, so that each cluster is associated with a certain classification; and
performing a pre-processing of the hyperspectral data, prior to at least one of determining the pixel-wise classification, performing the segmentation, or assigning the same classification.