Pneumoperitoneum refers to the presence of free air in the abdominal cav
ity. Accompanied with the clinical presentation of acute abdominal pain, p
neumoperitoneum found on imaging is highly suggestive of a perforated
viscus. Urgent surgical evaluation and intervention is required to reduce p
atient morbidity and mortality as delayed treatment can lead to septic sho
ck and multi-organ failure, eventually resulting in death.
Computed tomography (CT) is the best imaging modality in identifying p
neumoperitoneum. At present, response time to patients with pneumoper
itoneum is heavily dependent on the vigilance of attending clinicians and
turnaround reporting times of radiologists. This is however easily confoun
ded by congested emergency departments with long lists of imaging due
for reporting in addition to physician fatigue, amongst other factors.
We therefore embarked on the development of an artificial intelligence (A
I) algorithm via deep-learning to assist clinicians in the preliminary interpr
etation of CT abdominal imaging. Upon completion of CT scans, the imag
es are automatically uploaded into our AI algorithm for the screening of a
bnormal free gas in the abdominal cavity. The process takes 5 minutes, an
d those with positive findings are immediately alerted to the on-site radiol
ogists and/or physicians for further confirmation.
Our deep learning-powered novel algorithm and system thus assist with t
he rapid identification of pneumoperitoneum on abdominal CT imaging,
with the ultimate aim of hastening the subsequent surgical evaluation and
intervention required by these patients for better clinical outcomes.