Artificial Intelligence and the Role of Radiologists

A Deep Learning Approach for MRI in the Diagnosis of Labral Injuries of the Hip Joint

BACKGROUND

The diagnosis of labral injury on MRI is time-consuming and potential for incorrect diagnoses.

PURPOSE

To explore the feasibility of applying deep learning to diagnose and classify labral injuries with MRI.

STUDY TYPE

Retrospective.

POPULATION

A total of 1016 patients were divided into normal (n = 168, class 0) and abnormal labrum (n = 848) groups. The abnormal group consisted of n = 111 with class 1 (degeneration), n = 437 with class 2 (partial or complete tear), and n = 300 with unclassified injury. Patients were randomly divided into training, validation, and test cohort according to the ratio of 55%:15%:30%.

FIELD STRENGTH/SEQUENCE

Fat-saturation proton density-weighted fast spin-echo sequence at 3.0 T.

ASSESSMENT

Convolutional neural network-6 (CNN-6) was used to extract, discriminate, and detect oblique coronal (OCOR) and oblique sagittal (OSAG) images. Mask R-CNN was used for segmentation. LeNet-5 was used to diagnose and classify labral injuries. The weighting method combined the models of OCOR and OSAG. The output-input connection was used to correlate the whole diagnosis/classification system. Four radiologists performed subjective diagnoses to obtain the diagnosis results.

STATISTICAL TESTS

CNN-6 and LeNet-5 were evaluated by area under the receiver operating characteristic (ROC) curve and related parameters. The mean average precision (MAP) evaluated the Mask R-CNN. McNemar's test was used to compare the radiologists and models. A P value < 0.05 was considered statistically significant.

RESULTS

The area under the curve (AUC) of CNN-6 was 0.99 for extraction, discrimination, and detection. MAP values of Mask R-CNN for OCOR and OSAG image segmentation were 0.96 and 0.99. The accuracies of LeNet-5 in the diagnosis and classification were 0.94/0.94 (OCOR) and 0.92/0.91 (OSAG), respectively. The accuracy of the weighted models in the diagnosis and classification were 0.94 and 0.97, respectively. The accuracies of radiologists in the diagnosis and classification of labrum injuries ranged from 0.85 to 0.92 and 0.78 to 0.94, respectively.

DATA CONCLUSION

Deep learning can assist radiologists in diagnosing and classifying labrum injuries.

EVIDENCE LEVEL

3 TECHNICAL EFFICACY: Stage 2.

Clinical decision support system recommendations: how often do radiologists and clinicians accept them?

OBJECTIVE

To assess the acceptance and reliability of clinical decision support system (CDSS) imaging referral scores (ESR iGuide).

METHODS

A pilot study was conducted in a tertiary hospital. Four different experts were invited to rate 40 simulated clinical cases on a 5-level scale, for the level of agreement with the ESR iGuide's recommended procedures. In cases of disagreement, physicians were asked to indicate the reason. Descriptive measures were calculated for the level of agreement. We also explored the degree of agreement between four different specialists, and examined the cases in which clinicians disagreed with ESR iGuide best practice recommendations.

RESULTS

The mean rating of the four experts for the 40 clinical simulated cases was 4.17 ± 0.65, median 4.25 (on a scale of 1-5). All four raters totally agreed with the system recommendation in 75% of cases. No significant relationship was found between the degree of agreement and the number of indications and the patient's age or gender. In an optimistic scenario, using a binary agree/disagree variable, the Overall Percentage Agreement for the rating of the 40 simulated cases between the four experts was 77.28%. There were a total of 20 disagreements out of 160 cases with the ESR iGuide, of which 7 were among the two radiologists.

CONCLUSIONS

CDSS can be an effective tool for guiding the selection of appropriate imaging examinations, thus cutting costs due to unnecessary imaging scans. Since this is a pilot study, further research on a larger scale, preferably at national level, is required.

KEY POINTS

• The average of the mean rating of the four experts was 4.17 ± 0.65, median 4.25, on a scale of 1-5 where 5 represents total agreement with the CDSS tool. • In an optimistic scenario, using a binary agree/disagree variable, the Overall Percentage Agreement between the four experts was 77.28%. • Radiologists had fewer disagreements with the recommendations of the CDSS tool than other physicians, indicating a better fit of the support system to radiology experts' perspective.