A Proposed Machine Learning Based Collective Disease Model to Enable Predictive Diagnostics in Necrotising Enterocolitis

Abstract

Despite 60 years of research into necrotising enterocolitis (NEC), our understanding of the disease has not improved enough to achieve better outcomes. Even though NEC has remained the leading cause of death and poor outcomes in preterm infants, there remain vital questions on how to define, differentiate and detect the condition. Numerous international groups have recently highlighted NEC as a research priority and called for broader engagement of the scientific community to move the field forward. The three foremost barriers at present are lack of suitable definition(s), lack of clean datasets and consequently a lack of scope to gain sufficient insights from data. This research paper proposes a new direction of travel to advance neonatal gastro-intestinal monitoring and strengthen our efforts to gain better insights from global databases. An integrated machine learning based model is recommended to produce a comprehensive disease model to manage the complexity of this multi-variate disease. This intelligent disease model would be used in the daily neonatal settings to help aggregate data to support clinical decision making, better capture the complexity of each patient to enrich global datasets to create bigger and better data. This paper reviews current machine learning and CAD technologies in neonatology and suggests an innovative approach for an NEC disease model.

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

“…For medical NEC patients with multiple samples, a single AR image was randomly selected. In addition to ARs, clinical parameters previously reported in the literature [3], [7], [13], [15][16][17] were also extracted. A total of 23 clinical parameters that included demographic data (age, birth weight, gestational age, etc.…”

“…It is estimated that up to 50% of patients need surgical intervention, 46.5% of patients do not survive after surgery, and 20% to 50% of the survivors develop long-term sequelae, such as recurrence, intestinal stenosis, short bowel syndrome, slowed growth, and neurodevelopmental disorders [2]. NEC consists of a group of complex multivariable diseases that are difficult to describe, detect, and diagnose [3], [4]. Numerous international groups have recently highlighted NEC as a research priority and have made efforts to move the field forward [1], [5], [6].…”

“…Agakidou et al [4] showed that the prediction and diagnosis of NEC were satisfactory in the Era of Metabolomics and Proteomics, despite the small quantity of training data [4]. Additionally, van Druten et al [2], [3], [8] diagnosed NEC based on abdominal X-rays with artificial intelligence, but the results were not expected through a single medical examination. Multimodal models represent attempts to effectively simplify the complexity of multiple-factor diagnosis [9].…”

A Deep Learning Based Suggested Model to Detect Necrotising Enterocolitis in Abdominal Radiography Images

“…Neonatology and paediatrics lack computer-aided detection software (CAD) for neonatal imaging that can help screening, detection and ensuring standardised reporting of imaging findings [36], [37]. Furthermore, novel imaging modalities like AUS remains underutilised due to the lack of expertise in its use and reporting [36], [37]. Highlighting a further area for CAM to assist the integration of new point of care ultrasound (POCUS) to overcome underutilisation [26], [36], [37].…”

“…Furthermore, novel imaging modalities like AUS remains underutilised due to the lack of expertise in its use and reporting [36], [37]. Highlighting a further area for CAM to assist the integration of new point of care ultrasound (POCUS) to overcome underutilisation [26], [36], [37].…”