Sequential convolutional neural networks for nuclei segmentation
Methods and apparatus for segmenting cell nuclei in medical images apply first and second trained machine learning algorithms. The first trained machine learning algorithm processes a medical image to provide center locations of cell nuclei depicted in the image. The second machine learning algorithm processes each of a plurality of patches of the image. Each of the patches correspond to one of the plurality of center locations. Processing each patch yields a nuclear boundary corresponding to the corresponding one of the center locations. The methods and apparatus allow associating individual pixels of the image with one or more than one nuclei and have been shown to be effective for instance segmentation of nuclei in clusters of overlapping cell nuclei.
1 . A method for segmenting cell nuclei in medical images, the method comprising:
by a first trained machine learning algorithm processing a medical image to provide center locations of cell nuclei depicted in the medical image and outputting a plurality of patches in the medical image, each patch corresponding to one of the center locations and at least some of the patches overlapping one another in the medical image;
providing the plurality of patches output from the first trained machine learning algorithm as input to a second trained machine learning algorithm;
by the second trained machine learning algorithm processing the plurality of patches, the processing by the second trained machine learning algorithm outputting, for each of the plurality of patches, a nuclear boundary in the medical image corresponding to the corresponding one of the center locations, wherein at least a plurality of the nuclear boundaries overlap one another in the medical image; and
mapping each pixel of the medical image to one or more cell nuclei or no cell nuclei based on the nuclear boundaries, wherein at least some of the pixels are mapped to multiple cell nuclei.
2 . The method according to claim 1 wherein the first trained machine learning algorithm is implemented by a first convolutional neural network.
3 . The method according to claim 1 wherein processing the plurality of patches comprises receiving the plurality of patches as input to a second convolutional neural network.
4 . The method according to claim 3 wherein the first trained machine learning algorithm is implemented by a first convolutional neural network and at least one of the first and second convolutional neural networks has a UNet configuration.
5 . The method according to claim 1 wherein the plurality of patches is centered on the corresponding one of the center locations.
6 . The method according to claim 1 wherein the plurality of patches of a digital histopathology representation have dimensions of at least 80 by 80 pixels.
7 . The method according to claim 1 wherein the first trained machine learning algorithm is implemented by a first convolutional neural network, the second trained machine learning algorithm is implemented by a second convolutional neural network and the first and second convolutional neural networks have architectures that are one of different from one another and the same as one another.
8 . The method according to claim 1 further comprising obtaining cell information corresponding to the center locations and processing the cell information together with the center locations to perform cell type based cell-cell association quantification.
9 . The method according to claim 8 wherein the cell information comprises morphologically based and/or immunohistochemistry (IHC) based characterization information.
10 . The method according to claim 1 wherein the medical image comprises: a digital histopathology representation, a cytology image, a cytopathology image, or an in vivo histology image.
11 . The method according to claim 1 wherein the medical image includes one or more clusters of overlapping cell nuclei.
12 . The method according to claim 1 comprising applying feature calculations and a binary classification tree to classify objects corresponding to the nuclear boundaries.
13 . Apparatus for segmenting cell nuclei in medical images, the apparatus comprising a processor, the processor configured to:
by a first trained machine learning algorithm implemented by the processor, process a medical image to provide center locations of cell nuclei depicted in the medical image and output a plurality patches, each patch corresponding to one of the center locations and at least some of the patches overlapping one another in the medical image;
provide the plurality of patches output from the first trained machine learning algorithm as input to a second trained machine learning algorithm implemented by the processor;
by the second trained machine learning algorithm implemented by the processor, process the plurality of patches, the processing by the second trained machine learning algorithm outputting, for the plurality of patches, a nuclear boundary in the medical image corresponding to the corresponding one of the center locations, wherein at least a plurality of the nuclear boundaries overlap one another in the medical image; and
mapping each pixel of each medical image to one or more cell nuclei or no cell nuclei, wherein at least some of the pixels are mapped to multiple cell nuclei.
14 . The apparatus according to claim 13 wherein the first trained machine learning algorithm is implemented by a first convolutional neural network.
15 . The apparatus according to claim 13 wherein the second trained machine learning algorithm is configured to receive the plurality of patches as input to a second convolutional neural network.
16 . The apparatus according to claim 13 wherein the plurality of patches is centered on the corresponding one of the center locations.
17 . The apparatus according to claim 13 wherein the plurality of patches of a digital histopathology representation have dimensions of at least 80 by 80 pixels.
18 . The apparatus according to claim 13 wherein the first trained machine learning algorithm is implemented by a first convolutional neural network, the second trained machine learning algorithm is implemented by a second convolutional neural network and the first and second convolutional neural networks have architectures that are one of different from one another and the same as one another.
19 . The apparatus according to claim 13 wherein the processor is configured to obtain cell information corresponding to the center locations and to process the cell information together with the center locations to perform cell type based cell-cell association quantification.
20 . The apparatus according to claim 19 wherein the cell information comprises morphologically based and/or immunohistochemistry (IHC) based characterization information.
21 . The apparatus according to claim 13 wherein the medical image comprises: a digital histopathology representation, a cytology image, a cytopathology image, or an in vivo histology image.
22 . The apparatus according to claim 13 wherein the apparatus is operable to instance segment individual cell nuclei in one or more clusters of overlapping cell nuclei included in the medical image.
23 . The apparatus according to claim 13 wherein the processor is configured to apply one or more feature calculations and a binary classification tree to classify objects corresponding to the nuclear boundaries.