Hybrid 3D-to-2D slice-wise object localization ensembles

A computer-implemented method, comprising:

• accessing, by a device operatively coupled to a processor, at least one three-dimensional voxel array;
• and
• executing, by the device, a deep learning ensemble, to localize an object depicted in the at least one three-dimensional voxel array, wherein the deep learning ensemble:
• (i) applies hybrid three-dimensional and two-dimensional convolutional operations to the three-dimensional voxel array;
• (ii) performs the convolutional operations in a manner that maintains spatial correspondence of voxel data along a slicing axis of the three-dimensional voxel array by restricting down-sampling operations along the slicing axis;
• and
• (iii) generates, as a primary result of the executing, a plurality of two-dimensional object location indicators, each object location indicator being associated with a respective two-dimensional slice of the three-dimensional voxel array;
• wherein the deep learning ensemble comprises a neural network architecture including a backbone that applies three-dimensional convolutional kernels and a feature pyramid network that applies two-dimensional convolutional kernels on a slice-wise basis using shared weights.

A computer program product for facilitating hybrid 3D-to-2D slice-wise object localization ensembles, the computer program product comprising a computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:

• access at least one three-dimensional voxel array depicting a cervical spine of a medical patient;
• and
• localize, via execution of a deep learning ensemble, the program instructions cause the processor to:
• (i) perform hybrid three-dimensional and two-dimensional convolutional processing on the three-dimensional voxel array;
• (ii) process voxel data in a manner that preserves spatial correspondence along a slicing axis of the three-dimensional voxel array by restricting down-sampling operations along the slicing axis;
• and
• (iii) generate, as a primary localization result, a set of two-dimensional fracture location indicators, each fracture location indicator being indexed to a respective two-dimensional slice of the three-dimensional voxel array;
• wherein the deep learning ensemble comprises a neural network architecture including a backbone that applies three-dimensional convolutional kernels and a feature pyramid network that applies two-dimensional convolutional kernels on a slice-wise basis using shared weights.