Deep Learning-Based Algorithm for the Detection and Characterization of MRI Safety of Cardiac Implantable Electronic Devices on Chest Radiographs

Figures

• 

  Fig. 1. The flow chart of the study inclusion process. We refined the collected data in three steps. In the first step, the validity of each label was investigated and CRs with invalid labels were removed. Next, CRs with failure to convert DICOM files to PNG files were excluded. Last, CRs whose group did not include 20 instances were filtered out. Data set 1 is further split into three sets for model development: training, tuning, and test sets. BMC = SMG-SNU Boramae Medical Center, CR = chest radiograph, DICOM = digital imaging and communications in medicine, PNG = portable network graphics, SNUH = Seoul National University Hospital

• 

  Fig. 2. The data flow of XCID. The input CR passes through two main networks: the CCD and the MTL. The CCD locates the CIED region and the MTL identifies the manufacturer, model group, and MRI safety of the CIED. Before MTL receives the CIED region image, the CIED region is cropped, resized, and normalized. CCD = class-agnostic cardiac rhythm device detector, CIED = cardiac implantable electronic device, CR = chest radiograph, MTL = multi-task learner, XCID = X-ray cardiac implantable electronic device identification

• 

  Fig. 3. The confusion matrix of the multi-task learner for the model group identification task in the external test data set. Diagonal components not displayed in the confusion matrix indicate that the external test data set does not contain cardiac implantable electronic devices with those model groups.

• 

  Fig. 4. Representative CRs with successful classification and identification results. The right three columns show the original CRs overlaid with heatmaps. The multi-task learner reacts in three different manners for three tasks, resulting in three heatmaps for each CR. CR = chest radiograph

• 

  Fig. 5. Representative CRs with failure results of MTL. The CRs in the first two rows show obscured or distorted circuits in PGs by the over-angulated projection of PGs on CRs. The second to fourth rows were over-saturated radiographs that could not be used to classify PGs correctly. Overall, PGs with failed classification tended to present with overlapping ribs, scapula, pulmonary consolidations, and calcifications. The result in parenthesis indicates a subgroup classified by the MTL. CR = chest radiograph, MTL = multi-task learner, PG = pulse generator, Uncertain = under test or not approved as an MRI-compatible device