Relevance of pneumoperitoneum in the conservative approach to complicated acute diverticulitis. A retrospective study identifying risk factors associated with treatment failure

Nonoperative management of acute complicated diverticulitis with pericolic and/or distant extraluminal air: A systematic review

“…As 10 studies did not comply with inclusion criteria, only 19 underwent qualitative and quantitative synthesis. [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36][37][38] No additional records were found through other sources (references list).…”

“…The Newcastle-Ottawa Scale scores obtained ranged from 4 to 8. Specifically, five of the studies were deemed to be of poor quality (≤4 stars), 26,27,29,34,37 12 were of medium quality (5-6 stars), [20][21][22][23][24][25]28,30,31,33,35,36 and 2 were high-quality studies (7 or >7 stars). 32,38…”

“…Table 2 shows 19 included studies and their features. A total of 13 studies had a retrospective nature 20,21,23,24,26,28,[30][31][32][33][34]36,37 and 6 were prospective ones 22,25,27,29,35,38 ; 6 were multicentric [25][26][27]34,37,38 and 2 were comparative . 32,38 All identified studies were observational ones.…”

PACT-3D, a Deep Learning Algorithm for Pneumoperitoneum Detection in Abdominal CT Scans

“…This suggests that the model is more adept at identifying larger volumes of free air, particularly those originating from the upper gastrointestinal tract, where emergency surgical intervention is often imperative. Conversely, smaller volumes of free air, typically resulting from inflammatory conditions like acute diverticulitis, are usually managed with conservative treatment or elective surgery in patients who are hemodynamically stable 26 . These findings indicate that the PACT-3D model can also serve as a valuable tool for risk stratification by illustrating the perforated area alongside the volume of free air.…”

“…To augment the data, we normalized all CTs to 512×512×z-axis and randomly cubed them to 384×384×z-axis using the 'albumentations' library for each image in the training set 26 . The loss function we employed for the model combined Dice loss and Focal loss, each weighted at 50%.…”