Application of deep learning for medical imaging evaluation
This disclosure generally pertains to methods and systems for processing electronic data obtained from imaging or other diagnostic and evaluative medical procedures. Certain embodiments relate to methods for the development of deep learning algorithms that perform machine recognition of specific features and conditions in imaging and other medical data. Another embodiment provides systems configured to detect and localize medical abnormalities on medical imaging scans by a deep learning algorithm.
1. A method for streamlining a head scan interpretation workflow by automating an initial screening and triage process, comprising:
receiving and processing images from CT head scan imaging procedure data;
detecting and localizing medical abnormalities of the images using a deep learning system,
wherein the deep learning system carried out by a computer is developed by the steps of:
selecting medical imaging scans and extracting the medical abnormalities using natural language processing (NLP) algorithms, wherein each type of the medical abnormalities is annotated at scan, slice and pixel levels;
training a slice-wise deep learning algorithm with the selected medical imaging scans to segment pixel-level annotated scans;
training the deep learning algorithm with the selected medical imaging scans to generate slice-level confidences;
predicting confidences for a presence of each type of medical abnormalities; and
generating a score that corresponds to a level of recognition of the medical abnormalities and outputting a mask representing the precise location and extent of the medical abnormalities; and
prioritizing an assignment of a medical evaluation to an evaluator based on the score generated for the images.
2. The method of claim 1 , wherein the CT head scan imaging procedure data are transmitted to a Picture Archiving and Communication System (PACS) via Digital Imaging and Communications in Medicine (DICOM).
3. The method of claim 2 , wherein DICOM images are anonymized on a local server.
4. The method of claim 3 , wherein anonymized DICOM images are transferred to a cloud server.
5. The method of claim 4 , wherein detecting and localizing medical abnormalities of the images and prioritizing the assignment of a medical evaluation to an evaluator are carried out in the cloud server to obtain a prioritization status for DICOM images.
6. The method of claim 5 , wherein the prioritization status for DICOM images is transferred to the local server and a Health Level Seven (HL7) message is constructed in response to the prioritization status for DICOM images.
7. The method of claim 6 , wherein the HL7 message is transmitted to a Radiology Information System (RIS).
8. The method of claim 1 , wherein the deep learning algorithm comprises a convolutional neural network architecture.
9. The method of claim 8 , wherein the architecture is modified by using multiple parallel fully connected layers.
10. The method of claim 1 , wherein the confidences at the slice-level are combined across slices using a fully connected layer to predict a scan-level confidence for the presence and types of the medical abnormalities.
11. The method of claim 1 , wherein the medical abnormalities comprise intracranial hemorrhage and each of the 5 types of intracranial hemorrhages including intraparenchymal hemorrhage (IPH), intraventricular hemorrhage (IVH), subdural intracranial hemorrhage (SDH), epidural hemorrhage (EDH), and subarachnoid hemorrhage (SAH); midline shift; mass effect; skull fractures, and calvarial fractures.
12. The method of claim 1 , wherein the head CT scans are preprocessed by windowing to three separate windows.
13. The method of claim 12 , wherein the three separate windows comprise brain window, bone window and subdural window.
14. The method of claim 1 , wherein an accuracy of the deep learning algorithm for detecting the medical abnormalities is validated by comparing with radiologist reports.
15. The method of claim 1 , wherein the medical imaging procedure data comprise image data captured by an imaging modality and order data provided from a request for a radiological image read.
16. An automated head CT scan screening and triage system for streamlining a head scan interpretation workflow, comprising:
a deep learning system carried out by a computer to detect and localize medical abnormalities on non-contrast head CT scans, wherein the deep learning system is developed by the steps of:
selecting medical imaging scans and extracting medical abnormalities using natural language processing (NLP) algorithms, wherein each type of the medical abnormalities is annotated at scan, slice and pixel levels;
training a slice-wise deep learning algorithm with the selected medical imaging scans to segment pixel-level annotated scans;
training the deep learning algorithm with the selected medical imaging scans to generate slice-level confidences;
predicting confidences for a presence of each type of medical abnormalities; and
generating a score that corresponds to a level of recognition of the medical abnormalities and outputting a mask representing the precise location and extent of the medical abnormalities;
wherein the automated head CT scan screening and triage system is configured to receive and process images from CT head scan imaging procedure data;
wherein the automated head CT scan screening and triage system is configured to detect and localize the medical abnormalities of the images using the deep learning system; and
wherein the automated head CT scan screening and triage system is configured to prioritize an assignment of a medical evaluation to an evaluator based on the score generated for the images.
17. The system of claim 16 , wherein the CT head scan imaging procedure data are transmitted to a Picture Archiving and Communication System (PACS) via Digital Imaging and Communications in Medicine (DICOM).
18. The system of claim 17 , wherein DICOM images are anonymized on a local server.
19. The system of claim 18 , wherein anonymized DICOM images are transferred to a cloud server.
20. The system of claim 19 , wherein detecting and localizing medical abnormalities of the images and prioritizing the assignment of a medical evaluation to an evaluator are carried out in the cloud server to obtain a prioritization status for DICOM images.
21. The system of claim 20 , wherein the prioritization status for DICOM images is transferred to the local server and a Health Level Seven (HL7) message is constructed in response to the prioritization status for DICOM images.
22. The system of claim 21 , wherein the HL7 message is transmitted to a Radiology Information System (RIS).
23. The system of claim 16 , wherein the deep learning algorithm comprises a convolutional neural network architecture.
24. The system of claim 23 , wherein the architecture is modified by using multiple parallel fully connected layers.
25. The system of claim 16 , wherein the confidences at the slice-level are combined across slices using a fully connected layer to predict the scan-level confidence for the presence and types of the medical abnormalities.
26. The system of claim 16 , wherein the medical abnormalities comprise intracranial hemorrhage and each of the 5 types of intracranial hemorrhages including intraparenchymal hemorrhage (IPH), intraventricular hemorrhage (IVH), subdural intracranial hemorrhage (SDH), epidural hemorrhage (EDH), and subarachnoid hemorrhage (SAH); midline shift; mass effect; skull fractures and calvarial fractures.
27. The system of claim 16 , wherein the head CT scans are preprocessed by windowing to three separate windows.
28. The system of claim 27 , wherein the three separate windows comprise brain window, bone window and subdural window.
29. The system of claim 16 , wherein an accuracy of the deep learning algorithm for detecting the medical abnormalities is validated by comparing with the radiologist reports.
30. The system of claim 16 , wherein the medical imaging procedure data comprise image data captured by an imaging modality and order data provided from a request for a radiological image read.