Artificial neural network–based diagnostic models for lung cancer combining conventional indicators with tumor markers

Abstract

This study set out to establish a lung cancer diagnosis and prediction model uses conventional laboratory indicators combined with tumor markers, so as to help early screening and auxiliary diagnosis of lung cancer through a convenient, fast, and cheap way, and improve the early diagnosis rate of lung cancer. A total of 221 patients with lung cancer, 100 patients with benign pulmonary diseases, and 184 healthy subjects were retrospectively studied. General clinical data, the results of conventional laboratory indicators, and tumor markers were collected. Statistical Product and Service Solutions 26.0 was used for data analysis. The diagnosis and prediction model of lung cancer was established by artificial neural network – multilayer perceptron. After correlation and difference analysis, five comparison groups (lung cancer-benign lung disease group, lung cancer-health group, benign lung disease-health group, early-stage lung cancer-benign lung disease group, and early-stage lung cancer-health group) obtained 5, 28, 25, 16, and 25 valuable indicators for predicting lung cancer or benign lung disease, and then established five diagnostic prediction models, respectively. The area under the curve (AUC) of each combined diagnostic prediction model (0.848, 0.989, 0.949, 0.841, and 0.976) was higher than that of the diagnostic prediction model established only using tumor markers (0.799, 0.941, 0.830, 0.661, and 0.850), and the difference in the lung cancer-health group, the benign lung disease-health group, the early-stage lung cancer-benign lung disease group, and early-stage lung cancer-health group was statistically significant ( P < 0.05). The artificial neural network–based diagnostic models for lung cancer combining conventional indicators with tumor markers have high performance and clinical significance in assisting the diagnosis of early lung cancer.

Development and validation of a nomogram for predicting lung cancer based on acoustic-clinical features

“…Our results suggested that the nomogram model demonstrated high predictive accuracy, calibration, clinical applicability, and generalizability. At present, arti cial intelligence technology has demonstrated great potential and promising prospects, especially in the elds of lung cancer screening, imaging examination, pathological testing and biomarker detection [31] .Therefore, the potential of acoustic diagnosis for lung cancer remains largely unexplored. This study explores from the perspective of voice acoustic features for the rst time, combining it with clinical data and applying machine learning methods to construct models, laying the groundwork for intelligent diagnoses of lung cancer.…”

Utility of four machine learning approaches for identifying ulcerative colitis and Crohn's disease

“…Recent research has shown that ML algorithms are more advantageous than traditional statistical methods when constructing predictive models. Artificial neural networks are capable of self-learning, adaptation, fault-tolerance, nonlinearity, and efficient mapping of inputs to outputs [ 25 , 26 ], and have been used effectively to differentiate between mild cognitive impairment and Alzheimer's disease, to develop predictive models for lung cancer diagnosis, and for risk prediction of cardiovascular disease [ [27] , [28] , [29] ]. Decision trees are useful for determining groupings, recognizing connections between groups, and forecasting upcoming occurrences [ 30 ].…”