The broken circle method: a novel technique that enhances detection of Hill-Sachs lesions on internal rotation shoulder radiographs.

AIM: To describe and test a new method that increases the conspicuity of a Hill-Sachs lesion on internal rotation (IR) radiographs. MATERIALS AND METHODS: This study had institutional review board approval. A retrospective search for patients with a prior shoulder dislocation and a Hill-Sachs lesion documented on magnetic resonance imaging (MRI) was performed over a 10-year period identifying 256 test patients. In Part 1, the IR radiographs from test cases were randomised with controls, and three readers scored them independently for the defect. The readers were then taught the Broken Circle (BC) method and re-scored the radiographs. In Part 2, 15 cases of Hill-Sachs lesions that were missed by all readers in Part 1 were randomised with controls, and were shown to 25 radiology residents before (pre-test) and after (post-test) learning the BC method. A paired t-test was used to compare the differences in sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). RESULTS: In Part 1, the sensitivity increased 19.7% (54.1%-73.8%; p<0.05) and NPV increased 10.8% (62.5%-73.3%; p<0.01). In Part 2, post-test sensitivity for residents increased 16.3% (55.2%-71.5%; p<0.0001), accuracy increased 13.4% (64%-77.4%; p<0.0001), and NPV increased 13.3% (40.8%-54.1%; p<0.0001) independent of the level of training. The change in accuracy was also statistically significant for every individual class. CONCLUSION: The BC method was an effective technique that facilitated detection of a Hill-Sachs lesion at all levels of training, and was useful as a teaching tool.

Detection and localisation of hip fractures on anteroposterior radiographs with artificial intelligence: proof of concept.

AIM: To investigate the feasibility of applying a deep convolutional neural network (CNN) for detection/localisation of acute proximal femoral fractures (APFFs) on hip radiographs. MATERIALS AND METHODS: This study had institutional review board approval. Radiographs of 307 patients with APFFs and 310 normal patients were identified. A split ratio of 3/1/1 was used to create training, validation, and test datasets. To test the validity of the proposed model, a 20-fold cross-validation was performed. The anonymised images from the test cohort were shown to two groups of radiologists: musculoskeletal radiologists and diagnostic radiology residents. Each reader was asked to assess if there was a fracture and localise it if one was detected. The area under the receiver operator characteristics curve (AUC), sensitivity, and specificity were calculated for the CNN and readers. RESULTS: The mean AUC was 0.9944 with a standard deviation of 0.0036. Mean sensitivity and specificity for fracture detection was 97.1% (81.5/84) and 96.7% (118/122), respectively. There was good concordance with saliency maps for lesion identification, but sensitivity was lower for characterising location (subcapital/transcervical, 84.1%; basicervical/intertrochanteric, 77%; subtrochanteric, 20%). Musculoskeletal radiologists showed a sensitivity and specificity for fracture detection of 100% and 100% respectively, while residents showed 100% and 96.8%, respectively. For fracture localisation, the performance decreased slightly for human readers. CONCLUSION: The proposed CNN algorithm showed high accuracy for detection of APFFs, but the performance was lower for fracture localisation. Overall performance of the CNN was lower than that of radiologists, especially in localizing fracture location.

A database de-identification framework to enable direct queries on medical data for secondary use.

OBJECTIVE: To qualify the use of patient clinical records as non-human-subject for research purpose, electronic medical record data must be de-identified so there is minimum risk to protected health information exposure. This study demonstrated a robust framework for structured data de-identification that can be applied to any relational data source that needs to be de-identified. METHODS: Using a real world clinical data warehouse, a pilot implementation of limited subject areas were used to demonstrate and evaluate this new de-identification process. Query results and performances are compared between source and target system to validate data accuracy and usability. RESULTS: The combination of hashing, pseudonyms, and session dependent randomizer provides a rigorous de-identification framework to guard against 1) source identifier exposure; 2) internal data analyst manually linking to source identifiers; and 3) identifier cross-link among different researchers or multiple query sessions by the same researcher. In addition, a query rejection option is provided to refuse queries resulting in less than preset numbers of subjects and total records to prevent users from accidental subject identification due to low volume of data. This framework does not prevent subject re-identification based on prior knowledge and sequence of events. Also, it does not deal with medical free text de-identification, although text de-identification using natural language processing can be included due its modular design. CONCLUSION: We demonstrated a framework resulting in HIPAA Compliant databases that can be directly queried by researchers. This technique can be augmented to facilitate inter-institutional research data sharing through existing middleware such as caGrid.