Fig 1.
Example image of each modality and anatomical location.
(a) abdomen CT, (b) brain CT, (C) brain MRI, (d) MRI lumbar spine.
Fig 2.
Workflow chart.
Table 1.
Data distribution across the image modalities.
Fig 3.
Our CNN architecture.
Table 2.
Performance of our CNN architecture on image modality and anatomical location classification.
Table 3.
Multi-class confusion table for validation set of the five attempts.
Table 4.
Multi-class confusion table for testing set of the five attempts.
Fig 4.
Example of incorrect prediction by our proposed CNN architecture is shown.
(a) CT of brain image misclassified as MRI of brain image in the validation set. (b) MRI of spine image misclassified to CT of brain in the validation set. (c) MRI of spine image misclassified to MRI of brain in testing set. Notice the image in (a) contains intracerebral hemorrhage (arrow).
Table 5.
Overall performance measures of our CNN architecture on validation set (In percentage).
Table 6.
Overall performance measures of our CNN architecture on testing set (In percentage).
Table 7.
Summary of medical image classification on image modality and/or anatomy by various authors.