System and method for computer aided diagnosis
The present disclosure relates to a method for training a classifier. The method includes: acquiring an original image; determining a candidate target by segmenting the original image based on at least two segmentation models; determining a universal set of features by extracting features from the candidate target; determining a reference subset of features by selecting features from the universal set of features; and determining a classifier by performing classifier training based on the reference subset of features.
1. A system of computer aided diagnosis, comprising:
a message server duster distributing pending data, the pending data comprising an original image; and
a plurality of working nodes of a computer aided diagnosis server configured to:
(1) acquire the pending data from the message server cluster,
(2) segment the original image based on at least two segmentation models to determine a substance region of a candidate target and a surrounding region of the candidate target, and
(3) determine the candidate target based on the substance region and the surrounding region in a streaming parallel technique in real-time, wherein the determining a candidate target by segmenting the original image based on the at least two segmentation models comprises:
performing preliminary positioning on the original image to determine one or more first positioning regions,
performing, based on Hessian dot-enhancement, threshold segmentation on the one or more first positioning regions to determine a second positioning region, the second positioning region including the candidate target and a background region, the candidate target including a substance region and a surrounding region that surrounds the substance region,
determining the substance region of the candidate target based on the second positioning region and a statistical model,
determining the surrounding region of the candidate target based on the statistical model and a morphological model, and
combining the substance region and the surrounding region to determine the candidate target.
2. The system of claim 1 , wherein the message server cluster includes a Kafka cluster.
3. The system of claim 1 , wherein the pending data is processed by a Storm cluster.
4. The system of claim 1 , further comprising:
a duster management node configured to manage the message server duster and the plurality of working nodes, wherein the duster management node includes a Zookeeper duster.
5. The system of claim 1 , wherein the pending data include an original medical image or intermedia data, and the intermedia data include a region of interest (ROI), preprocessed data, a candidate target, or feature data.
6. A method of computer aided diagnosis, comprising:
managing a message server duster and a plurality of working nodes of a computer aided diagnosis server to perform operations including:
distributing pending data through the message server duster in a streaming mode, wherein the pending data includes an original image;
acquiring the pending data from the message server duster through the plurality of working nodes of the computer aided diagnosis server; and
processing the pending data in a streaming parallel mode in real time, the processing the pending data in a streaming parallel mode in real time including
(1) segmenting the original image based on at least two segmentation models to determine a substance region of a candidate target and a surrounding region of the candidate target, and
(2) determining the candidate target based on the substance region and the surrounding region in a streaming parallel technique in real-time, wherein the determining a candidate target by segmenting the original image based on the at least two segmentation models comprises:
performing preliminary positioning on the original image to determine one or more first positioning regions,
performing, based on Hessian dot-enhancement, threshold segmentation on the one or more first positioning regions to determine a second positioning region, the second positioning region including the candidate target and a background region, the candidate target including a substance region and a surrounding region that surrounds the substance region,
determining the substance region of the candidate target based on the second positioning region and a statistical model,
determining the surrounding region of the candidate target based on the statistical model and a morphological model, and
combining the substance region and the surrounding region to determine the candidate target.
7. A medical system, comprising:
an input device configured to acquire medical data;
a computer aided diagnosis device configured to acquire the medical data from the input device, the computer aided diagnosis device comprising:
a message server cluster configured to acquire the medical data from the input device and distribute the medical data, the medical data comprising an original image;
a plurality of working nodes of a computer aided diagnosis server configured to:
(1) acquire the medical data from the message server cluster,
(2) segment the original image based on at least two segmentation models to determine a substance region of a candidate target and a surrounding region of the candidate target, and
(3) determine the candidate target based on the substance region and the surrounding region in a streaming parallel technique in real-time,
wherein the determining a candidate target by segmenting the original image based on the at least two segmentation models comprises:
performing preliminary positioning on the original image to determine one or more first positioning regions,
performing, based on Hessian dot-enhancement, threshold segmentation on the one or more first positioning regions to determine a second positioning region, the second positioning region including the candidate target and a background region, the candidate target including a substance region and a surrounding region that surrounds the substance region,
determining the substance region of the candidate target based on the second positioning region and a statistical model,
determining the surrounding region of the candidate target based on the statistical model and a morphological model, and
combining the substance region and the surrounding region to determine the candidate target;
an output device configured to output the processing result.
8. The system of claim 7 , wherein the input device includes an imaging device configured to generate medical images, or a storage device configured to store data from the imaging device or data from the working nodes, the storage device includes at least one of a database, a PACS, or a file device.
9. The system of claim 7 , further comprising:
a cluster management node configured to manage a Kafka cluster and the plurality of working nodes, wherein the duster management node includes a Zookeeper cluster.