Method and apparatus for selecting radiology reports for image labeling by modality and anatomical region of interest
Systems and methods for developing a classification model for classifying medical reports, such as radiology reports. One method includes selecting, from a corpus of reports, a training set and a testing set, assigning labels of a modality and an anatomical focus to the reports in both sets, and extracting a sparse representation matrix for each set based on features in the training set. The method also includes learning, with one or more electronic processors, a correlation between the features of the training set and the corresponding labels using a machine learning classifier, thereby building a classification model and testing the classification model on the reports in the testing set for accuracy using the sparse representation matrix of the testing set. The method further includes predicting, with the classification model, labels of an anatomical focus and a modality for remaining reports in the corpus not included in the sets.
1. A method for developing a classification model, the method comprising:
selecting, from a corpus of reports, a subset of the reports from which to form a training set and a testing set;
assigning labels of a modality and an anatomical focus to the reports in both the training set and the testing set, wherein assigning labels of the modality and the anatomical focus to the reports including assigning a binary classification distinguishing chest x-ray reports from non-chest x-ray reports and, for each report labeled as a non-chest x-ray report, assigning a multiclass classification defining a modality associated with the report;
extracting a sparse representation matrix for each of the training set and the testing set based on features in the training set;
learning, with one or more electronic processors, a correlation between the features of the training set and the corresponding labels using a machine learning classifier, thereby building a classification model;
testing the classification model on the reports in the testing set for accuracy using the sparse representation matrix of the testing set; and
predicting, with the classification model, labels of an anatomical focus and a modality for remaining reports in the corpus not included in the subset.
2. The method of claim 1 , further comprising verifying the correctness of the predicted modality labels by:
(i) performing unsupervised feature clustering on the remaining reports, thereby forming corpus clusters;
(ii) measuring compactness of the corpus clusters formed;
(iii) performing unsupervised feature clustering on the reports in the training set, thereby forming training clusters; and
(iv) measuring overlap of the corpus clusters with the training clusters.
3. The method of claim 1 , wherein extracting the sparse representation matrix for each of the training set and the testing set based on features in the training set includes extracting the sparse representation matrix for each of the training set and the test set based on term frequency-inverse document frequency (TFIDF) features in the training set.
4. The method of claim 1 , wherein extracting the sparse representation matrix for each of the training set and the testing set based on features in the training set includes extracting the sparse representation matrix for each of the training set and the test set based on term frequency features in the training set.
5. The method of claim 1 , wherein learning the correlation between the features and their corresponding labels using the machine learning classifier includes learning the correlation using a logistic regression classifier.
6. The method of claim 1 , wherein learning the correlation between the features and their corresponding labels using the machine learning classifier includes learning the correlation using one selected from a group consisting of a logistic regression classifier, a decision tree classifier, and a support vector machine classifier.
7. The method of claim 1 , wherein predicting, with the classification model, the labels of the anatomical focus and the modality for the remaining reports includes classifying each of the remaining reports as one selected from a group consisting of mammography, chest X-ray, obstetric ultrasound, spine magnetic resonance imaging (MRI), spine X-ray, bone densitometry analysis (DEXA), chest, abdomen and pelvis computed tomography (cap CT), abdomen ultrasound, leg venous Doppler ultrasound, feet and ankle X-ray, positron-emission tomography PET/CT tumor imaging, chest computed tomography (CT), breast ultrasound, knee Mill, hip X-ray, knee X-ray, brain MM, breast MM, and thyroid ultrasound.
8. The method of claim 1 , wherein predicting, with the classification model, the labels of the anatomical focus and the modality for the remaining reports includes classifying each of the remaining reports as one selected from a group consisting of a chest X-ray report and a non-chest X-ray report.
9. A system for developing a classification model, the system comprising:
one or more electronic processors configured to:
select, from a corpus of reports, a subset of the reports from which to form a training set and a testing set,
assign labels of a modality and an anatomical focus to the reports in both the training set and the testing set, wherein assigning labels of the modality and the anatomical focus to the reports including assigning a binary classification distinguishing chest x-ray reports from non-chest x-ray reports and, for each report labeled as a non-chest x-ray report, assigning a multiclass classification defining a modality associated with the report,
extract a sparse representation matrix for each of the training set and the testing set based on features in the training set,
learn a correlation between the features of the training set and the corresponding labels using a machine learning classifier, thereby building a classification model,
test the classification model on the reports in the testing set for accuracy using the sparse representation matrix of the testing set, and
predict, with the classification model, labels of an anatomical focus and a modality for remaining reports in the corpus not included in the subset.
10. The system of claim 9 , wherein the one or more electronic processors are further configured to verifying a correctness of the predicted modality labels by:
(i) performing unsupervised feature clustering on the remaining reports, thereby forming corpus clusters;
(ii) measuring compactness of the corpus clusters formed;
(iii) performing unsupervised feature clustering on the reports in the training set, thereby forming training clusters; and
(iv) measuring overlap of the corpus clusters with the training clusters.
11. The system of claim 9 , wherein the features include frequency-inverse document frequency (TFIDF) features.
12. The system of claim 9 , wherein the features include term frequency features.
13. The system of claim 9 , wherein the machine learning classifier includes one selected from a group consisting of a logistic regression classifier, a decision tree classifier, and a support vector machine classifier.
14. The system of claim 9 , wherein the predicted labels of the anatomical focus and the modality for the remaining reports includes one selected from a group consisting of mammography, chest X-ray, obstetric ultrasound, spine magnetic resonance imaging (MM), spine X-ray, bone densitometry analysis (DEXA), chest, abdomen and pelvis computed tomography (cap CT), abdomen ultrasound, leg venous Doppler ultrasound, feet and ankle X-ray, positron-emission tomography PET/CT tumor imaging, chest computed tomography (chest CT), breast ultrasound, knee Mill, hip X-ray, knee X-ray, brain Mill, breast MM, and thyroid ultrasound.
15. The system of claim 9 , wherein the predicted labels of the anatomical focus and the modality for the remaining reports includes one selected from a group consisting of a chest X-ray report and a non-chest X-ray report.
16. A non-transitory, computer-readable medium storing instructions that, when executed by one or more electronic processors, perform a set of function, the set of functions comprising:
selecting, from a corpus of reports, a subset of the reports from which to form a training set and a testing set;
assigning labels of a modality and an anatomical focus to the reports in both the training set and the testing set, wherein assigning labels of the modality and the anatomical focus to the reports including assigning a binary classification distinguishing chest x-ray reports from non-chest x-ray reports and, for each report labeled as a non-chest x-ray report, assigning a multiclass classification defining a modality associated with the report;
extracting a sparse representation matrix for each of the training set and the testing set based on features in the training set;
learning a correlation between the features of the training set and the corresponding labels using a machine learning classifier, thereby building a classification model;
testing the classification model on the reports in the testing set for accuracy using the sparse representation matrix of the testing set; and
predicting, with the classification model, labels of an anatomical focus and a modality for remaining reports in the corpus not included in the subset.