From: WSES project on decision support systems based on artificial neural networks in emergency surgery
Authors | The training set size | Objectives of the research | Results |
---|---|---|---|
Acute appendicitis (AA) | |||
Yoldaş et al. [23] | 156 | AA diagnostics | Sensitivity—100%, specificity—97.2% |
Park and Kim [24] | 801 | AA diagnostics | The ANN proved to be more accurate in diagnosing AA (the accuracy of the three types of ANN—99.8%, 99.4%, 97.8%) than the Alvarado clinical scoring system (72.2%) |
Reismann et al. [25] | 590 | AA diagnostics, prediction of a complicated course of the disease in pediatrics | The ANN has allowed a significant improvement of the accuracy of diagnosis (sensitivity 93%, specificity 67%), and complicated course of AA (sensitivity 95%, specificity 33%) |
Park et al. [26] | 667 | AA diagnostics based on CT of patients with abdominal pains | The ANN showed good and very good diagnostic indicators of AA (accuracy > 90%) |
Acute pancreatitis (AP) | |||
Kazmierczak et al. [27] | 254 | Diagnosis of the AP by the level of pancreatic enzymes in the blood serum | Lipase level has the highest diagnostic accuracy (accuracy lipase—82%, serum amylase—76%, lipase + amylase—84%) |
Pofahl et al. [28] | 156 | Predicting the hospitalization length | Sensitivity 75%, specificity 81% and accuracy 79%, but the ANN predictive capabilities do not differ from Ranson and APACHE II |
Keogan et al. [29] | 92 | Predicting the hospitalization length based on CT and laboratory tests | The ANN showed the best predictive accuracy (AUC = 0.83 ± 0.05) compared to the Ranson (AUC = 0.68 ± 0.06; P < 0.02) and Balthazar (AUC = 0.62 ± 0.06; P < 0.003) |
Halonen et al. [30] | 234 | Predicting the potential mortality | The ANN predictive capabilities (AUC = 0.847) differ from Ranson (AUC = 0.655), APACHE II (AUC = 0.817) and Glasgow (AUC = 0.536) |
Mofidi et al. [31] | 496 | Identification of the AP severity and predicting lethal outcome | The ANN proved to be more accurate in diagnosing of the AP (ANN was more accurate than APACHE II and Glasgow in predicting: AP severity—P < 0.05 and P < 0.01 Multiple organ failure P < 0.05 and P < 0.01 Lethal outcome—P < 0.05 and P < 0.05) |
Andersson et al. [32] | 139 | Predicting the AP severity | The ANN proved to be more accurate (AUC = 0.92) in diagnosing of the severe AP in comparison with the logistic regression (AUC = 0.84, P = 0.03) and APACHE II (AUC = 0.63, P < 0.001) |
Hong et al. [33] | 312 | Predicting the persistent (more than 48 h) organ failure | The ANN proved to be more accurate in predicting of the persistent organ failure (AUC = 0.96 ± 0.02) in comparison with the logistic regression (AUC = 0.88 ± 0.03, P < 0.001) and APACHE II (AUC = 0.83 ± 0.03, P < 0.001) |
Fei et al. [34] | 152 | Predicting the severe AP associated with acute lung injury | The ANN proved to be more accurate (AUC = 0.859 ± 0.048) in predicting of the acute lung injury accompanying the AP in comparison with the logistic regression (AUC = 0.701 + 0.041) |
Acute cholecystitis (AC) | |||
Eldar et al. [35] | 180 | Predicting the conversion from laparoscopic to laparotomic access in AC | The ANN demonstrated a good predictive ability to predict the conversion from laparoscopic to laparotomic approach (100% of cases respectively, 67%—prospectively) and to determine the group of patients requiring laparotomic cholecystectomy |
Vukicevic et al. [36] | 303 | Prediction of choledocholithiasis in patients with gallstone disease and AC | The ANN demonstrated a good predictive ability to predict the choledocholithiasis and revealed informative clinical, laboratory, and instrumental signs (sensitivity—82.3%, specificity—94.7%, accuracy—92.2%, and AUC—0.934 in the validation set) |
Ulcerative bleeding | |||
Rotondano et al. [37] | 2380 | Predicting the fatal outcome in patients with bleeding from the upper gastrointestinal tract | The predictive ability of the ANN is better than the score Rockall’s one [38] (sensitivity—83.8% versus 71.4%, specificity—97.5% versus 52.0%, accuracy—96.8% versus 52.9%, and AUC—0.95 versus 0.67) |
Wong et al. [39] | 22,854 | Identification of patients with a high risk of recurrent bleeding requiring surgical treatment and with a high risk of death | The ANN demonstrated AUC = 0.78, and accuracy—84.3% |
Perforated gastroduodenal ulcers | |||
Søreide et al. [44] | 117 | Predicting the fatal outcome and determination of factors of the fatal outcome | AUC = 0.90, 0.95% CI [0.85–0.95] |
Ileus/bowel obstruction | |||
Cheng et al. [45] | 13,935 X-ray pictures | Ileus diagnostics | Sensitivity—91.4%, specificity—91.9% |
Strangulated hernia | |||
Chen et al. [46] | 762 | Predicting the need for bowel resection | ANN revealed eight factors that are significantly associated with the need for bowel resection |