Systems and methods for automated classification of a document

1. A computer-implemented method for extracting information from a dataset, comprising:

receiving, at an information handling device, a dataset;

extracting, via optical character recognition implemented by a processor of the information handling device, textual information associated with the dataset; and

classifying the dataset into one of a plurality of classes, the classifying further comprising:

computing a similarity score for each of the plurality of classes for each of a plurality of window regions of the dataset, the computing further comprising:

sliding a window across the textual information to define the plurality of window regions, and for each of the plurality of window regions:

computing a relevance metric for the window region; and

calculating the similarity score for each of the plurality of classes by calculating a similarity function between the relevance metric for the window region and an average relevance metric for each of the plurality of classes;

determining, based on a subset of highest similarity scores computed for each of the plurality of classes for each of the plurality of window regions, overall similarity scores for each of the plurality of classes for the dataset; and

classifying the dataset as corresponding to a class of the plurality of classes with a highest overall similarity score for the dataset.

2. The computer-implemented method of claim 1 , wherein the subset of highest similarity scores for each of the plurality of classes for each of the plurality of window regions are averaged to determine the overall similarity scores for each of the plurality of classes for the dataset.

3. The computer-implemented method of claim 1 , wherein the subset of highest similarity scores comprises one similarity score.

4. The computer-implemented method of claim 1 , wherein the similarity function is a cosine similarity.

5. The computer-implemented method of claim 1 , wherein computing the relevance metric includes computing a frequency-inverse dataset frequency (tf-idf) vector for the window region.

6. The computer-implemented method of claim 1 , the classifying further comprising:

determining whether the highest overall similarity score for the dataset meets or exceeds a threshold; and

if the highest overall similarity score for the dataset does not meet or exceed the threshold, classifying the dataset as corresponding to an indeterminate class.

7. The computer-implemented method of claim 1 , wherein the dataset is a medical report.

8. The computer-implemented method of claim 7 , wherein the dataset is a mammography report, wherein the textual information includes breast density information, and wherein the plurality of classes comprises at least one of: a fatty class, a scattered fibroglandular density class, a heterogeneously dense class, an extremely dense class, an indeterminate class, or a multiple class.

9. The computer-implemented method of claim 1 , further comprising:

preprocessing the dataset, the preprocessing including at least one of: removing at least one stopword from the dataset or stemming at least one word in the dataset.

10. The computer-implemented method of claim 9 , wherein the at least one stopword includes at least one of: it, what, is, are, the, a, an, and/or those.

11. The computer-implemented method of claim 1 , further comprising:

determining whether the dataset includes at least one predetermined keyword; and

wherein if the dataset is determined to include the at least one predetermined keyword, classifying the dataset into a class of the plurality of classes comprises only determining the class of the plurality of classes based on the at least one predetermined keyword, and

wherein the computing the similarity score for each of the plurality of classes for each of the plurality of window regions of the dataset, the determining the overall similarity scores for the dataset, and the classifying the dataset as corresponding to the class of the plurality of classes with a highest overall similarity score steps of the method of claim 1 are not performed.

12. The computer-implemented method of claim 11 , wherein the at least one predetermined keyword comprises at least one of: right, left, ipsilateral, contralateral, current, recent, previous, prior, history, “the breasts are almost entirely fat,” “the rest of the breasts are mildly dense,” “there are scattered densities throughout the breast,” “the breast tissue is dense,” “the breasts are heterogeneously dense, which may obscure small masses,” and/or “the right breast is very dense”.

13. The computer-implemented method of claim 1 , wherein a trained classifier performs the classifying the dataset into one of the plurality of classes.

14. The computer-implemented method of claim 13 , wherein the trained classifier is trained on a training set of datasets that are each already associated with a class of the plurality of classes, the training comprising:

computing a training relevance metric for each dataset of the training set of datasets; and

averaging the training relevance metric for all datasets of the training set that are associated with a same class of the plurality of classes to produce the average training relevance metric corresponding to each class of the plurality of classes.

15. The computer-implemented method of claim 14 , wherein the training set of datasets comprises a set of mammogram reports, wherein the associated class comprises at least one of: a fatty class, a scattered fibroglandular density class, a heterogeneously dense class, an extremely dense class, an indeterminate class, or a multiple class.

16. The computer-implemented method of claim 14 , the training further comprising:

for each dataset of the training set of datasets, extracting a region of text containing at least one focus word; and

training the trained classifier based on the region of text.

17. The computer-implemented method of claim 16 , wherein the at least one focus word comprises at least one of: dense, density, densities, heterogeneous, heterogeneously, scattered, fibroglandular, fat, fatty, extreme, extremely, moderate, mildly, largely, entirely, fatty replaced, average, and/or scattered areas of fibroglandular density.

18. The computer-implemented method of claim 13 , wherein the trained classifier comprises a set of hyperparameters, the set of hyperparameters including at least one of: a size of each sliding window region, a size of a focus region around a focus word, a number of highest similarity scores in the subset on which to base the overall similarity scores for the dataset, and a threshold.

19. The computer-implemented method of claim 18 , wherein the threshold indicates a minimum threshold similarity score.

20. The computer-implemented method of claim 14 , wherein the training further comprises optimizing a set of hyperparameters using an iterative grid search algorithm based on the training set of datasets.

21. The computer-implemented method of claim 20 , wherein the optimizing further comprises:

selecting, by an optimizer, a value for the set of hyperparameters;

calculating, by the optimizer, an accuracy of the trained classifier utilizing the set of hyperparameters; and

selecting, by the optimizer, an optimized value for the set of hyperparameters.

22. A computer system for extracting information from a dataset, the computer system comprising:

at least one memory storing instructions; and

at least one processor configured to execute the instructions to perform operations comprising:

access the at least one memory and execute processor-readable instructions, which when executed by the at least one processor configures the at least one processor to perform a plurality of functions, including functions for:

receiving, at an information handling device associated with the computer system, the dataset;

extracting, via optical character recognition implemented by the at least one processor, textual information associated with the dataset; and

classifying the dataset into one of a plurality of classes, the classifying further comprising:

computing a similarity score for each of the plurality of classes for each of a plurality of window regions of the dataset, the computing further comprising:

sliding a window across the textual information to define the plurality of window regions, and for each of the plurality of window regions:

 computing a relevance metric for the window region; and

 calculating the similarity score for each of the plurality of classes by calculating a similarity function between the relevance metric for the window region and an average relevance metric for each of the plurality of classes;

determining, based on a subset of highest similarity scores computed for each of the plurality of classes for each of the plurality of window regions, overall similarity scores for each of the plurality of classes for the dataset; and

classifying the dataset as corresponding to a class of the plurality of classes with a highest overall similarity score for the dataset.

23. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform operations for extracting information from a dataset, the operations comprising:

receiving the dataset;

extracting, via optical character recognition, textual information associated with the dataset; and

classifying the dataset into one of a plurality of classes, the classifying further comprising:

computing a similarity score for each of the plurality of classes for each of a plurality of window regions of the dataset, the computing further comprising:

sliding a window across the textual information to define the plurality of window regions, and for each of the plurality of window regions:

computing a relevance metric for the window region; and

calculating the similarity score for each of the plurality of classes by calculating a similarity function between the relevance metric for the window region and an average relevance metric for each of the plurality of classes;

determining, based on a subset of highest similarity scores computed for each of the plurality of classes for each of the plurality of window regions, overall similarity scores for each of the plurality of classes for the dataset; and

classifying the dataset as corresponding to a class of the plurality of classes with a highest overall similarity score for the dataset.

24. A computer-implemented method for extracting information from a dataset, comprising:

receiving, at an information handling device, the dataset;

extracting, via optical character recognition implemented by a processor of the information handling device, textual information associated with the dataset; and

classifying the dataset into one of a plurality of classes, the classifying further comprising:

computing a similarity score for each of the plurality of classes for each of a plurality of window regions of the dataset, wherein the computing includes sliding a window across the textual information to define the plurality of window regions;

determining, based on a subset of highest similarity scores, overall similarity scores for each of the plurality of classes for the dataset; and

classifying the dataset as corresponding to a class of the plurality of classes with a highest overall similarity score for the dataset.