Development of a Machine Learning-Based Screening Method for Thyroid Nodules Classification by Solving the Imbalance Challenge in Thyroid Nodules Data

Hematological indicator-based machine learning models for preoperative prediction of lymph node metastasis in cervical cancer

Background

Lymph node metastasis (LNM) is an important prognostic factor for cervical cancer (CC) and determines the treatment strategy. Hematological indicators have been reported as being useful biomarkers for the prognosis of a variety of cancers. This study aimed to evaluate the feasibility of machine learning models characterized by preoperative hematological indicators to predict the LNM status of CC patients before surgery.

Methods

The clinical data of 236 patients with pathologically confirmed CC were retrospectively analyzed at the Gynecology Oncology Department of the First Affiliated Hospital of Bengbu Medical University from November 2020 to August 2022. The least absolute shrinkage and selection operator (LASSO) was used to select 21 features from 35 hematological indicators and for the construction of 6 machine learning predictive models, including Adaptive Boosting (AdaBoost), Gaussian Naive Bayes (GNB), and Logistic Regression (LR), as well as Random Forest (RF), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost). Evaluation metrics of predictive models included the area under the receiver operating characteristic curve (AUC), accuracy, specificity, sensitivity, and F1-score.

Results

RF has the best overall predictive performance for ten-fold cross-validation in the training set. The specific performance indicators of RF were AUC (0.910, 95% confidence interval [CI]: 0.820-1.000), accuracy (0.831, 95% CI: 0.702-0.960), specificity (0.835, 95% CI: 0.708-0.962), sensitivity (0.831, 95% CI: 0.702-0.960), and F1-score (0.829, 95% CI: 0.696-0.962). RF had the highest AUC in the testing set (AUC = 0.854).

Conclusion

RF based on preoperative hematological indicators that are easily available in clinical practice showed superior performance in the preoperative prediction of CC LNM. However, investigations on larger external cohorts of patients are required for further validation of our findings.

Dual-layer detector spectral CT-based machine learning models in the differential diagnosis of solitary pulmonary nodules

The benign and malignant status of solitary pulmonary nodules (SPNs) is a key determinant of treatment decisions. The main objective of this study was to validate the efficacy of machine learning (ML) models featured with dual-layer detector spectral computed tomography (DLCT) parameters in identifying the benign and malignant status of SPNs. 250 patients with pathologically confirmed SPN were included in this study. 8 quantitative and 16 derived parameters were obtained based on the regions of interest of the lesions on the patients' DLCT chest enhancement images. 6 ML models were constructed from 10 parameters selected after combining the patients' clinical parameters, including gender, age, and smoking history. The logistic regression model showed the best diagnostic performance with an area under the receiver operating characteristic curve (AUC) of 0.812, accuracy of 0.813, sensitivity of 0.750 and specificity of 0.791 on the test set. The results suggest that the ML models based on DLCT parameters are superior to the traditional CT parameter models in identifying the benign and malignant nature of SPNs, and have greater potential for application.

Advancing accuracy in breath testing for lung cancer: strategies for improving diagnostic precision in imbalanced data

BACKGROUND

Breath testing using an electronic nose has been recognized as a promising new technique for the early detection of lung cancer. Imbalanced data are commonly observed in electronic nose studies, but methods to address them are rarely reported.

OBJECTIVE

The objectives of this study were to assess the accuracy of electronic nose screening for lung cancer with imbalanced learning and to select the best mechanical learning algorithm.

METHODS

We conducted a case‒control study that included patients with lung cancer and healthy controls and analyzed metabolites in exhaled breath using a carbon nanotube sensor array. The study used five machine learning algorithms to build predictive models and a synthetic minority oversampling technique to address imbalanced data. The diagnostic accuracy of lung cancer was assessed using pathology reports as the gold standard.

RESULTS

We enrolled 190 subjects between 2020 and 2023. A total of 155 subjects were used in the final analysis, which included 111 lung cancer patients and 44 healthy controls. We randomly divided samples into one training set, one internal validation set, and one external validation set. In the external validation set, the summary sensitivity was 0.88 (95% CI 0.84-0.91), the summary specificity was 1.00 (95% CI 0.85-1.00), the AUC was 0.96 (95% CI 0.94-0.98), the pAUC was 0.92 (95% CI 0.89-0.96), and the DOR was 207.62 (95% CI 24.62-924.64).

CONCLUSION

Electronic nose screening for lung cancer is highly accurate. The support vector machine algorithm is more suitable for analyzing chemical sensor data from electronic noses.

A Hybrid of Random Forests and Generalized Path Analysis: A Causal Modeling of Crashes in 52,524 Suburban Areas

BACKGROUND

Determining suburban area crashes' risk factors may allow for early and operative safety measures to find the main risk factors and moderating effects of crashes. Therefore, this paper has focused on a causal modeling framework.

STUDY DESIGN

A cross-sectional study.

METHODS

In this study, 52524 suburban crashes were investigated from 2015 to 2016. The hybrid-random-forest-generalized-path-analysis technique (HRF-gPath) was used to extract the main variables and identify mediators and moderators.

RESULTS

This study analyzed 42 explanatory variables using a RF model, and it was found that collision type, distinct, driver misconduct, speed, license, prior cause, plaque description, vehicle maneuver, vehicle type, lighting, passenger presence, seatbelt use, and land use were significant factors. Further analysis using g-Path demonstrated the mediating and predicting roles of collision type, vehicle type, seatbelt use, and driver misconduct. The modified model fitted the data well, with statistical significance ( χ230 =81.29, P<0.001) and high values for comparative-fit-index and Tucker-Lewis-index exceeding 0.9, as well as a low root-mean-square-error-of-approximation of 0.031 (90% confidence interval: 0.030-0.032).

CONCLUSION

The results of our study identified several significant variables, including collision type, vehicle type, seatbelt use, and driver misconduct, which played mediating and predicting roles. These findings provide valuable insights into the complex factors that contribute to collisions via a theoretical framework and can inform efforts to reduce their occurrence in the future.