Image-based detection of object condition
A system, method, and computer program product for image-based detection of conditions for an object internal to a patient is disclosed. A model can be trained for an internal object, such as an invasive medical device, the trained model being generated from one or more machine learning algorithms that are trained on annotated images of the object with condition information of the object. An imaging computer system can receive one or more images of the internal object captured by an imaging device positioned external to the patient. The imaging computer system can further determine, based on applying the trained model to the one or more images of the object, current condition information for the internal object, such as current spatial information for the object. A display can output the one or more images and the condition information for the object.
1. A system for image-based detection of an object internal to a patient, the system comprising:
a database configured to store a trained model, the trained model being generated from one or more machine learning algorithms that are trained on training images with annotated information identifying internal patient objects, wherein the trained model is configured to be used to detect an object from unannotated images;
an imaging computer system that is configured to:
receive, from an imaging device, one or more images of the object internal to a patient, wherein the imaging device is configured to capture the one or more images of the object from a position external to the patient,
access the trained model from the database,
detecting, based on applying the trained model to the one or more images of the object, the object internal to the patient from the one or more images, and
provide information identifying the object internal to the patient; and
a display to monitor the object internal to the patient, the display being configured to output (i) the one or more images of the object as captured by the imaging device from the position external to the patient, and (ii) the information identifying the object internal to the patient as determined based on application of the trained model to the one or more images.
2. The system of claim 1 , further comprising:
a training computer system to generate the trained model, the training computer system being programmed to:
obtain the training images of the internal patient objects with the annotated information, the training images depicting the internal patient objects, the annotated information identifying the internal patient objects,
iteratively train a model for the internal patient objects by correlating each of the training images to corresponding annotated information across one or more model layers using the one or more machine learning algorithms, wherein the iterative training generates the trained model for the internal patient objects, and
store the trained model for use by the imaging computer system.
3. The system of claim 2 , wherein at least a portion of the training images are actual images from use of the internal patient objects that have been manually annotated by a practitioner with the annotated information.
4. The system of claim 2 , wherein at least a portion of the training images are computer generated images simulating the internal patient objects as imaged by the imaging device.
5. The system of claim 2 , wherein the one or more machine learning algorithms comprises a supervised deep learning algorithm.
6. The system of claim 2 , wherein the trained model comprises a long short-term memory model.
7. The system of claim 1 , wherein the object comprises an invasive medical device.
8. The system of claim 7 , wherein the invasive medical device comprises a needle, a wire, a catheter, a probe, a stent, a drain, or a snare.
9. The system of claim 1 , wherein:
the system is further configured to monitor a plurality of objects internal to the patient,
the imaging computer system is further configured to detect the information identifying the plurality of objects, and
the display is further configured to output the information identifying the plurality of objects.
10. The system of claim 9 , wherein the plurality of objects include at least (i) a first object comprising an invasive medical device and (ii) a second object comprising an anatomical structure of the patient.
11. The system of claim 10 , wherein the information identifying the plurality of objects comprises spatial information for the first object and the second object relative to each other.
12. The system of claim 9 , wherein at least one of the plurality of objects comprises an anatomical structure of the patient.
13. The system of claim 1 , wherein the one or more images comprise one or more x-ray images, one or more ultrasound images, one or more computerized tomography (“CT”) scans, one or more magnetic resonance imaging (“MRI”) images, or one or more fluoroscopic images.
14. The system of claim 1 , wherein the information identifying the object comprises current spatial information for the object.
15. The system of claim 14 , wherein the current spatial information comprises orientation information that identifies a current orientation of the object internal to the patient.
16. The system of claim 14 , wherein the current spatial information comprises position information that identifies a current position of the object internal to the patient.
17. A method for image-based detection of an object internal to a patient, the method comprising:
receiving, from an imaging device, one or more images of the object internal to the patient, wherein the imaging device is configured to capture the one or more images of the object from a position external to the patient;
accessing a trained model from a database, wherein the trained model is generated from one or more machine learning algorithms trained on training images with annotated information identifying internal patient objects, wherein the trained model is configured to be used to detect the object from unannotated images;
detecting, based on applying the trained model to the one or more images of the object, the object internal to the patient; and
providing the information identifying the object for presentation on a display, the display being configured to output (i) the one or more images of the object as captured by the imaging device, and (ii) the information identifying of the object internal to the patient as determined based on application of the trained model to the one or more images.
18. The method of claim 17 , wherein the object comprises an invasive medical device.
19. The method of claim 17 , wherein the information identifying the object comprises current spatial information for the object.
20. A non-transitory computer readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
receiving, from an imaging device, one or more images of an object internal to the patient, wherein the imaging device is configured to capture the one or more images of the object from a position external to the patient;
accessing a trained model from a database, wherein the trained model is generated from one or more machine learning algorithms trained on training images with annotated information identifying internal patient objects, wherein the trained model is configured to be used to detect the object from unannotated images;
detecting, based on applying the trained model to the one or more images of the object, the object internal to the patient; and
providing the information identifying the object for presentation on a display, the display being configured to output (i) the one or more images of the object as captured by the imaging device, and (ii) the information identifying of the object internal to the patient as determined based on application of the trained model to the one or more images.
21. The system of claim 1 , wherein the object comprises an anatomical structure of the patient.
22. The system of claim 2 , wherein the one or more model layers comprise a convolutional layer, one or more pooling layers, and a plurality of residual layers.
23. The system of claim 22 , wherein applying the trained model to the one or more images of the object comprises, for each of the one or more images, feeding the image sequentially though (i) the convolutional layer, (ii) a first of the one or more pooling layer, (iii) each of the plurality of residual layers, and (iv) a second of the one or more pooling layers.
24. The system of claim 23 , wherein each of the plurality of residual layers comprises a subnetwork with one or more pathways of convolutional layers and a sum layer combining the outputs of the one or more pathways.
25. The system of claim 22 , wherein the convolutional layer comprises a K×K convolution with N filters to be applied at each input location to the convolution layer at one or more specified strides.
26. The method of claim 17 , wherein the object comprises an anatomical structure of the patient.
27. The method of claim 17 , wherein the one or more machine learning algorithms comprises a supervised deep learning algorithm and the trained model comprises a long short-term memory model.
28. The non-transitory computer readable medium of claim 20 , wherein the one or more machine learning algorithms comprises a supervised deep learning algorithm, and the trained model comprises a long short-term memory model.
29. The non-transitory computer readable medium of claim 20 , wherein the object comprises an invasive medical device.
30. The non-transitory computer readable medium of claim 20 , wherein the object comprises an anatomical structure of the patient.