Fig 1.
Left ear presents normal function of hearing. However, in right ear, there is a difference between air conductive (i.e., arrow) and bone conductive thresholds (i.e., circle). The audiogram for this patient suggests that there may be clinical issues related to the middle or external ear.
Table 1.
The table of SDS bin distribution on 12,697 subjects.
Table 2.
The hyper-parameters used in our proposed Random Forest model provided by scikit-learn framework.
Fig 2.
SDS prediction results using three models, multilayer perceptron, support vector machine, and Random Forest.
For training and testing of the three models, a dataset divided into 10 bins was used for training and evaluation. (a) Left ear, (b) Right ear.
Fig 3.
Confusion matrix for the SDS prediction performance using PTA thresholds.
The X-axis shows the predicted SDS and the Y-axis shows the ground-truth. The numbers in the matrix present the number of prediction occurrences. An ideal confusion matrix would have bright colors (high occurrence counts) diagonally. (a) Left ear, (b) Right ear.
Table 3.
Accuracy, precision, recall and F1 score of our SDS prediction model for the right and left ears.
Table 4.
The table of age distribution on 12,697 subjects.
Fig 4.
Confusion matrix for the age group prediction performance using PTA thresholds.
The X-axis shows the predicted age group of the patient and the Y-axis shows the ground-truth. The numbers in the matrix present the number of prediction occurrences. An ideal confusion matrix would have bright colors (high occurrence counts) diagonally. (a) Left ear, (b) Right ear.
Table 5.
Accuracy, precision, recall and F1 score of age prediction for the right and left ears.
Table 6.
Clinical remarks for subjects showing large bin differences (≥ 6) between predicted SDS and actual SDS.
Fig 5.
An average of pure tone audiograms (PTA) in all subjects’ AC and BC thresholds from 80-100 as shown in Fig 3.
(a) True positive on the left AC SDS. The ‘80-89’ SDS is correctly classified as ‘80-89’. (b) False negative on the left AC SDS. The ‘80-89’ SDS is misclassified as ‘90-100’. (c) True positive on the left BC SDS. The ‘80-89’ SDS is correctly classified as ‘80-89’. (d) False negative on the left BC SDS. The ‘80-89’ SDS is misclassified as ‘90-100’. (e) True positive on the left AC SDS. The ‘90-100’ SDS is correctly classified as ‘90-100’. (f) False negative on the left AC SDS. The ‘90-100’ SDS is misclassified as ‘80-89’. (g) True positive on the left BC SDS. The ‘90-100’ SDS is correctly classified as ‘90-100’. (h) False negative on the left BC SDS. The ‘90-100’ SDS is misclassified as ‘80-89’.
Fig 6.
Confusion matrix for the SDS prediction performance with a range of 0-90 only.
The X-axis shows the predicted SDS and the Y-axis shows the ground-truth. The numbers in the matrix present the number of prediction occurrences. An ideal confusion matrix would have bright colors (high occurrence counts) diagonally. (a) Left ear, (b) Right ear.