Multimodal AI System for the Rapid Diagnosis and Surgical Prediction of Necrotizing Enterocolitis

Abstract

The rapid diagnosis and surgical prediction of necrotizing enterocolitis (NEC) remain a challenge because its complex pathogenesis has not been completely elucidated, and no single medical examination is specific for diagnosing NEC. Artificial intelligence (AI) has proven the robustness of multivariate analysis and been widely used in the diagnosis of complex diseases in the past decade. In this paper, a new multimodal AI system including feature engineering, machine learning (ML), and deep learning (DL) was constructed based on abdominal radiographs (ARs) and clinical data. A total of 4,535 ARs from 1,823 suspected NEC patients were analyzed by transfer learning, and then medical images and clinical parameters from 827 suspected NEC patients were used to train, validate, and test the AI system. Our results demonstrated that the system was effective in diagnosing NEC. In addition, the clinical datasets obtained one week before surgery from 379 NEC patients were studied by the multimodal AI system, and the results showed that it was capable of predicting which NEC patients required surgery. We compared the results in external test sets with those made by clinicians and found that the diagnostic and surgical predictive ability of the AI system was equivalent to that of experienced clinicians. This multimodal AI system can help clinicians improve diagnostic efficiency, reduce the number of missed diagnoses, and facilitate early diagnosis and treatment to prevent disease progression or even death.

The risk prediction models of neonatal necrotizing enterocolitis in China: a systematic review

“…A total of 12 NEC risk prediction models were established in the 10 included literature, with a focus on premature infants and mostly single center studies,the total sample size is 177-1173.In terms of model building methods, ten models [9,[11][12][13][14][15][16]18] only used logistic regression algorithm, one model [10] used four algorithms: logistic regression, support vector machine, multilayer perceptron, and extreme gradient boosting, and one model [17] used three algorithms: artificial neural network, decision tree, and extreme gradient boosting.The modeling is detailed in Table 3.…”

Knowledge atlas on the medical artificial intelligence research: a bibliometric analysis based on CiteSpace and VOSviewer

“…Gao W [68] built a new multimode AI system including feature engineering, machine learning and deep learning based on abdominal X-ray pictures and clinical data, which can help clinicians improve diagnosis efficiency, reduce missed diagnosis times, promote early diagnosis and treatment, and prevent disease progression and even death. Wang X [69] put forward a prediction model based on optimization algorithm and neural network, which can select and sort the most important factors affecting the mental health of medical staff, predict the mental health status of medical staff around the world, and help to make appropriate work plans for medical staff.…”

Framework for feature selection of predicting the diagnosis and prognosis of necrotizing enterocolitis

“…As an example, let n = 20 and filp = 5, where n denotes the number of features in Refsol. If the index of features are 0 to 19, feature subsets f 0 , f 1 , f 2 , f 3 and f 4 are obtained by flipping the following bits, as shown in Fig 4: (0,5,10,15), (1,6,11,16), (2,7,12,17), (3,8,13,18) and (4,9,14,19). In the second strategy, the k-th subset of features is obtained by flipping n/flip contiguous bits starting from the k-th bit.…”

“…In the second strategy, the k-th subset of features is obtained by flipping n/flip contiguous bits starting from the k-th bit. Following the previous example, the feature subsets f 0 , f 1 , f 2 , f 3 and f 4 are obtained by flipping the following bits: (0,1,2,3), (4,5,6,7), (8,9,10,11), (12,13,14,15) and (16,17,18,19). With the above two searching strategies, we determine the search area for each bee.…”

“…In the feature selection stage, they use statistical analysis to extract important features. In the classification stage, most researchers use ML methods such as linear discriminant analysis (LDA) [14,15], random forest (RF) [16][17][18], or Light Gradient Boosting Machine (GBM) [19] as classifier models. Table 1 summarizes some studies using ML for the diagnosis or prognosis of NEC.…”