Enhanced pneumothorax visualization in ICU patients using portable chest radiography

Abstract

ObjectivePneumothorax development can cause precipitous deterioration in ICU patients, therefore quick and accurate detection is vital. Portable chest radiography is commonly performed to exclude pneumothoraces but is hampered by supine patient position and overlying internal and external material. Also, the initial evaluation of the chest radiograph may be performed by a relatively inexperienced physician. Therefore, a tool that could significantly improve pneumothorax detection on portable radiography would be helpful in patient care. The aim of this study was to evaluate the clinical utility of novel enhancement software for pneumothorax detection in readers with varied clinical experience of detecting/excluding pneumothoraces on portable chest radiographs in ICU patients.Subjects and methods206 portable ICU chest radiographs, 103 with pneumothoraces, were processed with and without enhancement software and reviewed by 5 readers who varied in reading experience. Images were grouped for different complexity levels.ResultsThe mean AUC for pneumothorax detection increased for 4/5 readers from 0.846–0.957 to 0.88–0.971 with a largest improvement for the reader with least experience. No significant change was noted for the reader with the longest reading experience. The image complexity had no impact on the interpretation result.ConclusionPneumothorax detection improves with novel enhancement software; the largest improvement is seen in less experienced readers.

Comparison of Baseline, Bone-Subtracted, and Enhanced Chest Radiographs for Detection of Pneumothorax

“…Another limitation of our study is the cost of the artificial intelligence (AI) algorithm, added interpretation time (for reviewing additional images), and need for additional training of radiologists for using the AI algorithm. 9,10 Finally, as opposed to the portable CXR used in the prior study on pneumothorax detection on bone-subtracted CXRs, 7 the machine learning–based bone subtraction in our study is compatible with non-portable, frontal CXRs only.…”

“…Ley-Zaporozhan et al have also reported improved detection of pneumothorax in 206 enhanced, portable chest radiographs (103 CXR-positive and 103 CXR-negative for pneumothorax) with a greater AUC (0.88-0.97) for the least experienced readers as compared to the AUC on unprocessed CXRs (0.85-0.96). 7 Since enhancement increases the conspicuity of ''edges'' or makes them ''sharper,'' it helps in the detection of pneumothorax.…”

“…Other studies have reported on the enhancement of CXR with image post-processing to improve detection of pneumothorax relative to unprocessed CXR. 7 While the software-based enhancement is a cheaper option than the hardware-based dual-energy radiography, it can increase the interpretation time for review of additionally enhanced images for assessing pneumothorax or the position of lines and tubes. Furthermore, enhanced images can mask pulmonary findings such as consolidation or edema.…”

“…Furthermore, enhanced images can mask pulmonary findings such as consolidation or edema. 7 The single-energy, machine learning-based, bonesubtraction technique used in our study suppresses ribs and clavicles on CXR by using advanced post-processing techniques without increasing the radiation dose or using specific imaging hardware. 3 While the dual-energy technique requires a change in the acquisition protocol for CXR, the single-energy, bone subtraction technique works on the image domain.…”

Improvement in Lung Insufflation in Spontaneously Breathing Tracheostomized Patients by Using a New Pulmonary Expansion Device: A Pilot Study

“…Transactions of the ASME (absence of disease) [26,27]. The perception of dyspnea was assessed using the Borg scale and was analyzed by comparing pre-and post-therapy scores.…”