Artificial intelligence supported reading by redacting of a normal area in a medical image
For supporting reading for radiologists, artificial intelligence (AI) identifies normal anatomy or parts of the patient represented in a medical image. Those parts are then altered, such as redacted, to indicate that no review of that part is needed. The reading presents medical images for review with one or more parts identified as not needing review (e.g., by being missing through redaction).
1. A method for artificial intelligence supported reading in a medical system, the method comprising:
obtaining, by a medical imager, a medical image representing a patient, wherein a portion of the medical image comprises at least a first anatomical part of the patient;
determining, by a first artificial intelligence examination of the medical image, that the first anatomical part of the patient is normal;
redacting the portion of the medical image comprising the first anatomical part of the patient from the medical image; and
displaying, during a reading by a radiologist, the medical image with the first anatomical part redacted wherein a remaining portion of the medical image is unredacted.
2. The method of claim 1 wherein obtaining the medical image comprises obtaining the medical image as a three-dimensional representation as the medical image, wherein displaying during the reading comprises displaying of multiple two-dimensional images of the three-dimensional representation, at least two of the two-dimensional images having the first anatomical part redacted.
3. The method of claim 1 wherein determining comprises determining that an organ as the first anatomical part is normal.
4. The method of claim 1 wherein determining comprises determining, by the first artificial intelligence, that a defect or disease is not in the medical image, a lack of the defect or disease indicating the normal.
5. The method of claim 4 wherein the first artificial intelligence was trained to detect the defect or disease for the first anatomical part, and a second artificial intelligence was trained to detect a different defect or disease for the first anatomical part, and wherein the first anatomical part is determined as normal only when both the first and second artificial intelligences fail to detect the defect or disease and the different defect or disease.
6. The method of claim 1 wherein determining comprises classifying the first anatomical part as normal.
7. The method of claim 1 wherein redacting comprises masking the first anatomical part as a blank region in the medical image.
8. The method of claim 1 wherein redacting comprises altering the first anatomical part in the medical image.
9. The method of claim 1 further comprising:
detecting, by a second artificial intelligence, a plurality of parts of the patient represented in the medical image, the plurality of parts including the first anatomical part and a second anatomical part; and
determining, by a third artificial intelligence examination of the medical image, that the second anatomical part of the patient is normal;
wherein redacting comprises redacting both the first and the second anatomical parts from the medical image.
10. The method of claim 1 further comprising segmenting the first anatomical part, and wherein determining comprises determining based on input of the first anatomical part as segmented to the first artificial intelligence.
11. A method for supported reading in a medical system, the method comprising:
detecting, by a machine-learned detector, different anatomies represented in scan data of a patient;
determining, by different machine-learned classifiers, whether the different anatomies are normal, different ones of the different machine-learned classifiers determining for respective different ones of the different anatomies;
altering a representation of at least one of the different anatomies when the at least one of the different anatomies is normal; and
displaying a medical image of the representation as altered, the medical image being generated from the scan data.
12. The method of claim 11 wherein multiple of the different machine-learned classifiers are applied to one of the different anatomies, the multiple of the different machine-learned classifiers having been trained to detect different diseases or abnormalities for the one of the different anatomies.
13. The method of claim 11 wherein displaying comprises displaying the medical image comprises displaying as part of a radiologist reading.
14. The method of claim 11 wherein the different machine-learned classifiers are selected from a library of machine-learned classifiers based on the anatomies detected as being represented in the scan data.
15. The method of claim 14 further comprising: altering the machine-learned classifiers available in the library.
16. The method of claim 11 wherein altering comprises redacting the representation.
17. A medical system comprising:
a memory configured to store a medical image of a patient and a first machine-learned normality classifier wherein the memory is configured to store a second machine-learned normality classifier and a third machine-learned normality classifier for different parts of the patient represented in the medical image;
an image processor configured to apply the first machine-learned normality classifier to a part of a patient represented in the medical image and to redact the part from the medical image when the first machine-learned normality classifier determines that the part is normal, wherein the image processor is further configured to apply the second and third machine-learned normality classifiers to the different parts and to redact the different parts from the medical image when the second and third machine-learned normality classifiers determine that the different parts are normal; and
a display configured to display the medical image with the normal part and normal different parts redacted.
18. The medical system of claim 17 wherein the first machine-learned normality classifier comprises a classifier for a first abnormality or disease of the part, wherein the memory is configured to store a second machine-learned normality classifier for a second abnormality or disease of the part, wherein the image processor is configured to apply the second machine-learned normality classifier to the part and to redact the part from the medical image when the first machine-learned normality classifier and the second machine-learned normality classifier both determine that the part is normal.
19. The medical system of claim 17 wherein the memory is configured to store a machine-learned detector, wherein the image processor is configured to apply the machine-learned detector to the medical image, the machine-learned detector configured to detect the part, the image processor configured to select the first machine-learned normality classifier based on the detection of the part.