Method and system for approximating deep neural networks for anatomical object detection

A method for anatomical object detection in a medical image comprising:

• training a deep neural network to detect the anatomical object in medical images;
• calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network;
• and
• detecting the anatomical object in a received medical image of a patient using the approximation of the trained deep neural network, wherein calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network comprises:
• reconstructing a respective trained weight matrix for each of a plurality of nodes in a current layer of the trained deep neural network using 1-D Haar wavelet bases and respective wavelet coefficients and reducing a number of wavelet coefficients used to reconstruct each respective trained weight matrix;
• and
• re-training the approximation of the trained deep neural network resulting from reconstructing the respective trained weight matric for each of the plurality of nodes in the current layer of the trained deep neural network and reducing the number of the wavelet coefficients used to reconstruct each respective trained weight matrix.

A method for anatomical object detection in a medical image comprising:

• training a deep neural network to detect the anatomical object in medical images;
• calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network;
• and
• detecting the anatomical object in a received medical image of a patient using the approximation of the trained deep neural network, wherein calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network comprises:
• for each of a plurality of layers in the trained deep neural network, applying principal component analysis (PCA) to a space of trained weight matrices over all of a plurality of nodes in that layer.

An apparatus for anatomical object detection in a medical image comprising:

• means for training a deep neural network to detect the anatomical object in medical images;
• means for calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network;
• and
• means for detecting the anatomical object in a received medical image of a patient using the approximation of the trained deep neural network, wherein the means for calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network comprises:
• means for reconstructing a respective trained weight matrix for each of a plurality of nodes in each of a plurality of layers of the trained deep neural network using 1-D Haar wavelet bases and wavelet coefficients.

A non-transitory computer readable medium storing computer program instructions for anatomical object detection in a medical image, the computer program instructions when executed by a processor cause the processor to perform operations comprising:

• training a deep neural network to detect the anatomical object in medical images;
• calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network;
• and
• detecting the anatomical object in a received medical image of a patient using the approximation of the trained deep neural network, wherein calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network comprises:
• for each of a plurality of nodes in each of a plurality of layers of the trained deep neural network, reconstructing a trained weight matrix for the node using 1-D Haar wavelet bases and wavelet coefficients.

A non-transitory computer readable medium storing computer program instructions for anatomical object detection in a medical image, the computer program instructions when executed by a processor cause the processor to perform operations comprising:

• training a deep neural network to detect the anatomical object in medical images;
• calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network;
• and
• detecting the anatomical object in a received medical image of a patient using the approximation of the trained deep neural network, wherein calculating an approximation of the trained deep neural network that reduces the computational complexity of the trained deep neural network comprises:
• for each of a plurality of layers in the trained deep neural network, applying principal component analysis (PCA) to a space of trained weight matrices over all of a plurality of nodes in that layer.