Table 1 Characteristics of studies in artificial intelligence applications for appendicitis diagnosis
Study, year | Objective(s) | Algorithm(s) applied | Sample size (training: validation) | Age, mean ± SD (year) | Sex (male/female) |
|---|---|---|---|---|---|
Park et al. [1], 2023 (South Korea) | Develop a CNN model for automated classification of AA | CNN | 715 cases in total (4078 CT images) 246 cases had AA (1959 CT images) Total cases (568:147) AA cases (199:47) | Total patients: 44.3 ± 18.4 AA: 41.9 ± 19.2 | Total patients: 368/347 AA: 130/116 |
Akbulut et al. [2], 2023 (Turkey) | Develop an ML model for the classification of AA and predicting perforated and non-perforated AA | CatBoost | 1797 (80%/20%) (AA: 1465, NA: 332, AA non-perforated 1161, AA perforated: 304) | Male (median: 33; IQR: 23). female (median: 34; IQR: 26). AA (median: 33.1; IQR: 25), NA (median: 33; IQR: 24) | 993/804 |
Ghareeb et al. [3], 2021 (Egypt) | Determine how well an AI-based model performs in diagnosing AA compared to using the Alvarado score alone or in combination with ultrasound criteria | Subspace KNN, KNN, LR, DT, SVM, NB | 319 (224: 95) | 30.5 ± 12.5 | 47.6%/52.4% |
Rajpurkar et al. [4], 2020 (USA) | Develop a 3D DL model called “AppendiXNet” for appendicitis detection using a small training dataset of under 500 CT examinations | CNN | Training: 438 examinations (255 appendicitis and 183 NA from 435 patients), Development: 106 examinations (53 appendicitis and 53 NA from 105 patients) Test: 102 examinations (51 appendicitis and 51 NA from 102 patients) | Training: 38.2 ± 15.6 development: 39.2 ± 17.3 Test: 38.4 ± 15.7 | Training: 182/253 Developmental: 41/64 Test: 38/64 |
Park et al. [5], 2020 (USA) | Examine the viability of a CNN-based diagnostic algorithm for AA using abdominopelvic CT scans | CNN | Training and internal validation: 667 (215 AA, 452 NA), External validation: Institution 1: 60 patients (26 AA, 34 NA), Institution 2: 40 patients (20 AA, 20 NA) | Training and internal validation: 45.6 ± 22.2 External validation: Institution1 (45.9 ± 18.9), Institution2 (43.9 ± 20.8) | Training and internal validation: 331/336, External validation: Institution 1 (25/35), Institution 2 (24/16) |
Zhao et al. [6], 2020 (China) | Provide potential urinary markers and an efficient model for AA diagnosis | Naïve Bayes, SVM, and RF | 134: 73/61 (48 patients had AA and 86 patients had other acute abdomens) AA (32 training: 16 validation) NA (41 training: 45 validation) 495 healthy controls | AA (training: 38:5 ± 17:9, validation: 34:2 ± 15:2), NA (training: 55:1 ± 18:2, validation: 56:6 ± 17:2) | AA (training: 14/18, validation: 9/7), NA (training: 13/28, validation: 14/17) |
Ramirez garcialunaa et al. [7], 2020 (Mexico) | Evaluate skin IRT imaging as a diagnostic adjunct for AA in adults | RF | 122: 98/24 (71 patients: 51 had AA, 20 patients had other diagnoses. 51 healthy controls) | AA: 29.1 ± 14.1, other patients: 30.0 ± 11.8, Healthy controls: 21.0 ± 3.3 | AA: 25/26, Other patients: 5/15, Healthy control: 31/20 |
Kang et al. [8], 2019 (South Korea) | Assess the effectiveness of a clinical approach using DT for diagnosis and compare its diagnostic performance with existing scoring systems | DT | 244 (80 patients with AA, 164 patients with other diagnoses) | AA: 35 (23–42) NA: 31 (23–41) | AA: 31/49 NA: 54/109 |
Gudelis et al. [9], 2019 (Spain) | Develop a diagnostic model for RIF pain. Specifically, the study focuses on creating a diagnosis model based on classification trees using the CHAID method and an ANN | ANN (MLP + BP), DT | 252 (93 patients had AA) | Total: 33.3 ± 16 AA: 37 ± 17 | Total: 52.8%/47.2% AA: 74.2%/25.6% |
Shahmoradi et al. [10], 2018 (Iran) | Construct a model to predict AA based on pathology reports | Radial Basis Function Network (RBFN), MLP, and LR | 181 cases (133 patients had AA) | Average: 28 | 126/55 |
Jamshidnezhad et al. [11], 2017 (Iran) | Develop a diagnostic model using minimal clinical factors within the initial hours of abdominal pain | Fuzzy-rule-based system (using Honey Bee Reproduction Cycle (HRBC)) | 70%/30% | NR | NR |
Park et al. [12], 2015 (South Korea) | Propose an appendicitis diagnosis system using ANN | RBFNN, MLNN, PNN | 801 cases I. NA (N = 596) II. Appendicitis (N = 205, No AA (N = 143), AA (n = 62)) | Total: 30.27 ± 18.58, NA: 29.68 ± 13.63, No AA: 30.55 ± 13.99 AA: 31.53 ± 16.32 | NA: 290/306, No AA: 53/90, AA: 32/30 |
Safavi et al. [13], 2015 (Iran) | Compare the ANN models and conventional laboratory tests in the diagnosis of appendicitis | ANN (MLP) | 100 (83 AA, 17 NA): training: 60, validation: 15, testing: 25 | 28/01 ± 12/68 | 71/29 |
Lee et al. [14], 2013 (Southern Taiwan) | Evaluate the prediction effectiveness of the PEL technique, addressing imbalanced sample learning issues, to support accurate diagnosis of AA | PEL | 574 (464 positive appendicitis, 110 negative appendicitis) | 36.18 (3–87) Positive appendicitis: 36.97 Negative appendicitis: 32.5 | 323/251 |
Yoldaş et al. [15], 2012 (Turkey) | Create a diagnostic model with ANNs and assess its effectiveness in diagnosing AA | ANN | 156 (132 appendicitis, 24 NA) | Total: 29.9 ± 10.8, Appendicitis: 29.3 ± 10.6, NA: 33.2 ± 11.9 | Total: 79/77, Appendicitis: 72/60, NA: 7/17 |
Sun et al. [16], 2012 (South Korea) | Build a hybrid decision support model to accurately diagnose suspected AA and identify useful decision rules by combining statistical analysis and DT algorithms | DT | 326 cases (90%:10%): 152 AA, 174 NA | AA: 36.57 ± 21.31 NA: 43.05 ± 20.86 | AA: 77/75 NA: 66/108 |
Hsieh et al. [17], 2011 (Taiwan) | Evaluate the performance of RF, SVM, and ANN in diagnosing AA | RF, SVM, ANN, LR | 180 (135:45), 115 patients had appendicitis | 39.4 (16–85) | 85/95 |
Ting et al. [18], 2010 (Taiwan) | Modify the ASS with decision tree technology and construct a convenient and accurate decision support model for AA diagnosis and timing of laparotomy | DT | 532 (340 patients had AA, 80 had perforated appendicitis, 112 NA) | AA: 31.9 NA: 29.9 Ruptured appendicitis: 37.1 | Total: 327/205 |
Prabhudesai et al. [19], 2008 (UK) | Assess the use of ANN for diagnosing appendicitis in patients with acute right iliac fossa pain | MLP type ANN | 60 patients with suspected appendicitis: 24 had appendicitis, 36 had other diagnosis | 25.4 | 27/33 |
Sakai et al. [20], 2007 (Japan) | Compare the diagnostic accuracy levels of ANN models and LR models for diagnosing AA | ANN and LR | 169 (86 cases had AA, 83 cases NA) | AA: 24.4 ± 20.3, NA: 27.5 ± 17.4 | 77/92 AA: 42/44, NA: 35/48 |
Pesonen et al. [21], 1996 (Finland) | Compare four ANN algorithms for diagnosing AA | Four different types of NN: two unsupervised learning and feedback networks including binary ART1 and Kohonen SOM. LVQ (supervised learning and feedback network), and BP (supervised learning and feedforward network) | 911 (454:457) | NR | 47.7%/52.3% |
Forsstrom et al. [22], 1995 (Finland) | Create the DIAGAID software package to establish a SMARTTP link connecting patient databases and clinicians | Neuro-fuzzy systems (DiagaiD), BPNN, LR | 186 (120:66), 145 patients had AA, and 41 did not have) | NR | NR |