Techniques for medical image retrieval
View Patent ↗A technique for searching for an image includes calculating wavelet features of a plurality of images. A keyword included in radiographic interpretation information is extracted for each of the stored images. The calculated wavelet features and the extracted keywords are stored in association with the respective stored images. A newly taken image is acquired and a wavelet feature of the acquired image is calculated. A keyword included in radiographic interpretation information corresponding to the acquired image is extracted and a search for similar radiographic interpretation information from the stored keywords is performed. A wavelet feature-based spatial distance between the acquired image and each of images corresponding to the radiographic interpretation information found is calculated. A search result of any images for which the calculated wavelet feature-based spatial distance is shorter than a predetermined value is output, in ascending order of the calculated wavelet feature-based spatial distance.
1. A medical image search apparatus for searching for an image similar to a newly taken image from among medical images based on past cases, comprising:
a feature calculation unit configured to calculate wavelet features of a plurality of images that have been taken and stored in the past;
a keyword extraction unit configured to extract a keyword included in radiographic interpretation information for each stored image;
an information storage unit configured to store the calculated wavelet features and the extracted keywords in association with the respective stored images;
an image acquisition unit configured to acquire a newly taken image;
a wavelet feature calculation unit configured to calculate a wavelet feature of the acquired image;
a radiographic interpretation information search unit configured to extract a keyword included in radiographic interpretation information corresponding to the acquired image and, on the basis of the extracted keyword, search for similar radiographic interpretation information from the stored keywords;
a spatial distance calculation unit configured to calculate a wavelet feature-based spatial distance between the acquired image and each of images corresponding to the radiographic interpretation information found; and
an output unit configured to output, as a search result, any images for which the calculated wavelet feature-based spatial distance is shorter than a predetermined value, in ascending order of the calculated wavelet feature-based spatial distance.
2. The apparatus according to claim 1 , wherein the feature calculation unit and the wavelet feature calculation unit each calculate two-dimensional Gabor wavelet features as the wavelet features.
3. The apparatus according to claim 1 , further comprising:
a frequency distribution vector calculation unit configured to calculate frequency distribution vectors for all images, by calculating M said wavelet features for each image and binarize the respective wavelet features for conversion into an M-dimensional bit string, wherein the spatial distance calculation unit calculates the calculated wavelet feature-based spatial distance as an angle between the calculated frequency distribution vectors, and wherein M is a natural number of 2 or greater.
4. A computer program product, comprising:
a computer-readable storage device; and
program code embodied on the computer-readable storage device, wherein the program code, when executed by a processor, configures the processor to:
calculate wavelet features of a plurality of images that have been taken and stored in the past;
extract a keyword included in radiographic interpretation information for each stored image;
store the calculated wavelet features and the extracted keywords in association with the respective stored images;
acquire a newly taken image;
calculate a wavelet feature of the acquired image;
extract a keyword included in radiographic interpretation information corresponding to the acquired image and, on the basis of the extracted keyword, search for similar radiographic interpretation information from the stored keywords;
calculate a wavelet feature-based spatial distance between the acquired image and each of images corresponding to the radiographic interpretation information found; and
output, as a search result, any images for which the calculated wavelet feature-based spatial distance is shorter than a predetermined value, in ascending order of the calculated wavelet feature-based spatial distance.
5. The computer program of claim 4 , wherein the wavelet features are two-dimensional Gabor wavelet features.
6. The computer program of claim 4 , wherein the program code, when executed by the processor, further causes the processor to:
calculate frequency distribution vectors for all images by calculating M said wavelet features for each image and binarize the respective wavelet features for conversion into an M-dimensional bit string, wherein M is a natural number of 2 or greater, and
calculate the calculated wavelet feature-based spatial distance as an angle between the calculated frequency distribution vectors.