Neonatal necrotizing enterocolitis, a serious gastrointestinal problem that inflames intestinal tissue causing it to die, is a severe disease that affects newborns. For this sensitive group, timely detection of a need for surgery is crucial in order to determine the best time for treatment and to improve prognosis. Researchers from the Zhejiang Children’s Hospital and Shanghai Maritime University collaborated in order to establish an algorithm model based on multimodal data to identify the characteristics of surgical indications and construct a diagnosis model.
The suggested algorithm uses Joint Nonnegative Matrix Factorization (JNMF) to mine the correlations between the data. In addition, to prevent overfitting, the adjacency matrix is employed as a network regularization constraint. In order to avoid feature redundancy and perform feature selection, orthogonal and L1-norm regulations were introduced, and confirmed 14 clinical features. Finally, to perform the classification of patients requiring surgery, researchers employed three classifiers: random forest, support vector machine and logistic regression. The results show that, when features selected by the proposed algorithm model are classified by random forest, the area under the ROC curve is 0.8, which has high prediction accuracy.
建議的算法使用聯合非負矩陣分解 (JNMF) 來挖掘數據之間的相關性。此外，為了防止過擬合，鄰接矩陣被用作網絡正則化約束。為了避免特徵冗餘和進行特徵選擇，引入了正交和 L1 範數規則，並確認了 14 個臨床特徵。最後，為了對需要手術的患者進行分類，研究人員使用了三種分類器：隨機森林、支持向量機和邏輯回歸。結果表明，本文算法模型選取的特徵進行隨機森林分類時，ROC曲線下面積為0.8，具有較高的預測精度。
Reference: Qi G, Huang S, Lai D, Li J, Zhao Y, Shen C, Huang J, Liu T, Wei K, Dou J, Shu Q, Yu G. Integrating abdominal plain radiographs and clinical data for neonatal necrotizing enterocolitis using hypergraph-based multi-constraint combined non-negative matrix factorization to construct a diagnostic model. Bosn J of Basic Med Sci [Internet]. 2022May15 [cited 2022May21];. Available from: https://www.bjbms.org/ojs/index.php/bjbms/article/view/7046
Editor: Merima Bukva, MPharm