Development and Validation of Machine Learning-based Model for the Prediction of Malignancy in Multiple Pulmonary Nodules: Analysis from Multicentric Cohorts

Deep Learning Models for Predicting Malignancy Risk in CT-Detected Pulmonary Nodules: A Systematic Review and Meta-analysis

BACKGROUND

There has been growing interest in using artificial intelligence/deep learning (DL) to help diagnose prevalent diseases earlier. In this study we sought to survey the landscape of externally validated DL-based computer-aided diagnostic (CADx) models, and assess their diagnostic performance for predicting the risk of malignancy in computed tomography (CT)-detected pulmonary nodules.

METHODS

An electronic search was performed in four databases (from inception to 10 August 2023). Studies were eligible if they were peer-reviewed experimental or observational articles comparing the diagnostic performance of externally validated DL-based CADx models with models widely used in clinical practice to predict the risk of malignancy. A bivariate random-effect approach for the meta-analysis on the included studies was used.

RESULTS

Seventeen studies were included, comprising 8553 participants and 9884 nodules. Pooled analyses showed DL-based CADx models were 11.6% more sensitive than physician judgement alone, and 14.5% more than clinical risk models alone. They had a similar pooled specificity to physician judgement alone [0.77 (95% CI 0.68-0.84) v 0.81 (95% CI 0.71-0.88)], and were 7.4% more specific than clinical risk models alone. They had superior pooled areas under the receiver operating curve (AUC), with relative pooled AUCs of 1.03 (95% CI 1.00-1.07) and 1.10 (95% CI 1.07-1.13) versus physician judgement and clinical risk models alone, respectively.

CONCLUSION

DL-based models are already used in clinical practice in certain settings for nodule management. Our results show their diagnostic performance potentially justifies wider, more routine deployment alongside experienced physician readers to help inform multidisciplinary team decision-making.

Preoperative localization for pulmonary nodules: a meta-analysis of coil and liquid materials

PURPOSE

This study was designed to conduct pooled comparisons of the relative clinical efficacy and safety of computed tomography (CT)-guided localization for pulmonary nodules (PNs) using either coil- or liquid material-based approaches.

MATERIAL AND METHODS

Relevant articles published as of July 2023 were identified in the Web of Science, PubMed, and Wanfang databases, and pooled analyses of relevant endpoints were then conducted.

RESULTS

Six articles that enrolled 287 patients (341 PNs) and 247 patients (301 PNs) that had respectively undergone CT-guided localization procedures using coil- and liquid material-based approaches prior to video-assisted thoracic surgery (VATS) were included in this meta-analysis. The liquid material group exhibited a significantly higher pooled successful localization rate as compared to the coil group (p = 0.01), together with significantly lower pooled total complication rates (p = 0.0008) and pneumothorax rates (p = 0.01). Both groups exhibited similar rates of pulmonary hemorrhage (p = 0.44) and successful wedge resection (p = 0.26). Liquid-based localization was also associated with significant reductions in pooled localization and VATS procedure durations (p = 0.004 and 0.007).

CONCLUSIONS

These data are consistent with CT-guided localization procedures performed using liquid materials being safer and more efficacious than coil-based localization in patients with PNs prior to VATS resection.

Unenhanced computed tomography radiomics help detect endoleaks after endovascular repair of abdominal aortic aneurysm

OBJECTIVES

To explore the feasibility of unenhanced CT images for endoleak detection of abdominal aortic aneurysm (AAA) after endovascular repair (EVAR).

METHODS

Patients who visited our hospital after EVAR from July 2014 to September 2021 were retrospectively collected. Two radiologists evaluated the presence or absence of endoleaks using the combination of contrast-enhanced and unenhanced CT as the referenced standard. After segmenting the aneurysm sac of the unenhanced CT, the radiomic features were automatically extracted from the region of interest. Histogram features of patients with and without endoleak were statistically analyzed to explore the differences between the two groups. Twelve common machine learning (ML) models based on radiomic features were constructed to evaluate the performance of endoleak detection with unenhanced CT images.

RESULTS

The study included 216 patients (69 ± 8 years; 191 men) with AAA, including 64 patients with endoleaks. A total of 1955 radiomic features of unenhanced CT were extracted. Compared with patients without endoleak, the aneurysm sac outside the stent of patients with endoleak had higher CT attenuation (41.7 vs. 33.6, p < 0.001) with smaller dispersion (51.5 vs. 58.8, p < 0.001). The average area under the curve (AUC) of the ML models constructed with unenhanced CT radiomics was 0.86 ± 0.05, the accuracy was 81% ± 4, the sensitivity was 88% ± 10, and the specificity was 78% ± 5. When fixing the sensitivity to > 90% (92% ± 2), the models retained specificity at 72% ± 10.

CONCLUSIONS

Unenhanced CT features exhibit significant differences between patients with and without endoleak and can help detect endoleaks in AAA after EVAR with high sensitivity.

CLINICAL RELEVANCE STATEMENT

Unenhanced CT radiomics can help provide an alternative method of endoleak detection in patients who have adverse reactions to contrast media. This study further exploits the value of unenhanced CT examinations in the clinical management and surveillance of postoperative abdominal aortic aneurysm.

KEY POINTS

• Unenhanced CT features of the aneurysm sac outside the stent exhibit significant differences between patients with and without endoleak. The endoleak group showed higher unenhanced CT attenuation (41.7 vs 33.6, p < .001) with smaller dispersion (51.5 vs 58.8, p < .001) than the nonendoleak group. • Unenhanced CT radiomics can help detect endoleaks after intervention. The average area under the curve (AUC) of twelve common machine learning models constructed with unenhanced CT radiomics was 0.86 ± 0.05, the average accuracy was 81% ± 4. • When fixing the sensitivity to > 90% (92% ± 2), the machine learning models retained average specificity at 72% ± 10.

Development and validation of machine learning-based model for mortality prediction in patients with acute basilar artery occlusion receiving endovascular treatment: multicentric cohort analysis

BACKGROUND

Predicting mortality in stroke patients using information available before endovascular treatment (EVT) is an essential component for supporting clinical decision-making. Although the mortality rate of acute basilar artery occlusion (ABAO) after EVT has reached 40%, few studies have focused on predicting mortality in these individuals. Thus, we aimed to develop and validate a machine learning-based mortality prediction tool based on preoperative information for ABAO patients receiving EVT.

METHODS

The derivation cohort comprised patients from southern provinces of China in the BASILAR registry. The model (POSITIVE: Predicting mOrtality of baSilar artery occlusion patIents Treated wIth EVT) was trained and optimized using a fivefold cross-validation method in which hyperparameters were selected and fine-tuned. This model was retrospectively tested in patients from the northern provinces of China from the BASILAR registry. A prospective test of POSITIVE was performed on consecutive patients from two hospitals between January 2020 and June 2022.

RESULTS

Extreme gradient boosting was employed to construct the POSITIVE model, which achieved the best predictive performance among the eight machine learning algorithms and showed excellent discrimination (area under the curve (AUC) 0.83, 95% confidence interval (95% CI) 0.80 to 0.87) and calibration (Hosmer-Lemeshow test, P>0.05) in the development cohort. AUC yielded by the POSITIVE model for the retrospective test was 0.79 (95% CI 0.71 to 0.85), higher than that obtained by traditional models. Prospective comparisons showed that the POSITIVE model achieved the highest AUC (0.82, 95% CI 0.74 to 0.90) among all prediction models.

CONCLUSION

We developed a machine learning algorithm and retrospective and prospective testing with multicentric cohorts, which exhibited a solid predictive performance and may act as a convenient reference to guide decision-making for ABAO patients. The POSITIVE model is presented online for user-friendly access.

Diagnosis of primary lung cancer and benign pulmonary nodules: a comparison of the breath test and 18F-FDG PET-CT

With the application of low-dose computed tomography in lung cancer screening, pulmonary nodules have become increasingly detected. Accurate discrimination between primary lung cancer and benign nodules poses a significant clinical challenge. This study aimed to investigate the viability of exhaled breath as a diagnostic tool for pulmonary nodules and compare the breath test with 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET)-computed tomography (CT). Exhaled breath was collected by Tedlar bags and analyzed by high-pressure photon ionization time-of-flight mass spectrometry (HPPI-TOFMS). A retrospective cohort (n = 100) and a prospective cohort (n = 63) of patients with pulmonary nodules were established. In the validation cohort, the breath test achieved an area under the receiver operating characteristic curve (AUC) of 0.872 (95% CI 0.760-0.983) and a combination of 16 volatile organic compounds achieved an AUC of 0.744 (95% CI 0.7586-0.901). For PET-CT, the SUVmax alone had an AUC of 0.608 (95% CI 0.433-0.784) while after combining with CT image features, 18F-FDG PET-CT had an AUC of 0.821 (95% CI 0.662-0.979). Overall, the study demonstrated the efficacy of a breath test utilizing HPPI-TOFMS for discriminating lung cancer from benign pulmonary nodules. Furthermore, the accuracy achieved by the exhaled breath test was comparable with 18F-FDG PET-CT.

Behavioural activity pattern, genetic factors, and the risk of nonalcoholic fatty liver disease: A prospective study in the UK Biobank

BACKGROUND & AIMS

Physical activity, sedentary behaviour, and genetic variants have been associated with the nonalcoholic fatty liver disease (NAFLD). However, whether and how the degree of healthy activity patterns may modify the impact of genetic susceptibility on NAFLD remains unknown.

METHODS

Behaviour activity factors were determined according to total physical activity (TPA) and sedentary time. The polygenic risk score (PRS) was calculated by variants in PNPLA3, TM6SF2, MBOAT7, and GCKR. Cox regression was used to analyse the associations of genetic and behaviour activity factors with incident NAFLD in the UK Biobank (N = 338 087).

RESULTS

During a median follow-up of 12.4 years, 3201 incident NAFLD cases were ascertained. Analyses of TPA and sedentary time simultaneously showed a dose-response association with the risk of NAFLD (ptrend  < .001). The association of behaviour activity patterns with NAFLD varied by genetic variants. Of the subjects with high genetic risk, we observed a null protective effect of moderate or high TPA on NAFLD risk, while sitting less than three hours a day significantly decreased the risk of NAFLD (p = 3.50 × 10-4 ). The high genetic risk of NAFLD can also be offset by the combination of moderate physical activity and shorter sedentary time. Moreover, the high genetic risk group has the greatest reduction of 10-year absolute risk (6.95 per 1000 person-years) if reaching both healthy activities.

CONCLUSIONS

Moderate-to-high physical activity and favourable sedentary behaviour may be lifestyle modifications in preventing NAFLD, which could offset the harmful effect of predisposing genetic factors.

Integration of IDPC Clustering Analysis and Interpretable Machine Learning for Survival Risk Prediction of Patients with ESCC

Precise forecasting of survival risk plays a pivotal role in comprehending and predicting the prognosis of patients afflicted with esophageal squamous cell carcinoma (ESCC). The existing methods have the problems of insufficient fitting ability and poor interpretability. To address this issue, this work proposes a novel interpretable survival risk prediction method for ESCC patients based on extreme gradient boosting improved by whale optimization algorithm (WOA-XGBoost) and shapley additive explanations (SHAP). Given the imbalanced nature of the data set, the adaptive synthetic sampling (ADASYN) is first used to generate the samples with high survival risk. Then, an improved clustering by fast search and find of density peaks (IDPC) algorithm based on cosine distance and K nearest neighbors is used to cluster the patients. Next, the prediction model for each cluster is obtained by WOA-XGBoost and the constructed model is visualized with SHAP to uncover the factors hidden in the structured model and improve the interpretability of the black-box model. Finally, the effectiveness of the proposed scheme is demonstrated by analyzing the data collected from the First Affiliated Hospital of Zhengzhou University. The results of the analysis reveal that the proposed methodology exhibits superior performance, as indicated by the area under the receiver operating characteristic curve (AUROC) of 0.918 and accuracy of 0.881.

Multi-omics integrated circulating cell-free DNA genomic signatures enhanced the diagnostic performance of early-stage lung cancer and postoperative minimal residual disease

BACKGROUND

Liquid biopsy is a promising non-invasive alternative for cancer screening and minimal residual disease (MRD) detection, although there are some concerns regarding its clinical applications. We aimed to develop an accurate detection platform based on liquid biopsy for both cancer screening and MRD detection in patients with lung cancer (LC), which is also applicable to clinical use.

METHODS

We applied a modified whole-genome sequencing (WGS) -based High-performance Infrastructure For MultIomics (HIFI) method for LC screening and postoperative MRD detection by combining the hyper-co-methylated read approach and the circulating single-molecule amplification and resequencing technology (cSMART2.0).

FINDINGS

For early screening of LC, the LC score model was constructed using the support vector machine, which showed sensitivity (51.8%) at high specificity (96.3%) and achieved an AUC of 0.912 in the validation set prospectively enrolled from multiple centers. The screening model achieved detection efficiency with an AUC of 0.906 in patients with lung adenocarcinoma and outperformed other clinical models in solid nodule cohort. When applied the HIFI model to real social population, a negative predictive value (NPV) of 99.92% was achieved in Chinese population. Additionally, the MRD detection rate improved significantly by combining results from WGS and cSMART2.0, with sensitivity of 73.7% at specificity of 97.3%.

INTERPRETATION

In conclusion, the HIFI method is promising for diagnosis and postoperative monitoring of LC.

FUNDING

This study was supported by CAMS Innovation Fund for Medical Sciences, Chinese Academy of Medical Sciences, National Natural Science Foundation of China, Beijing Natural Science Foundation and Peking University People's Hospital.

Real-world comprehensive diagnosis and "Surgery + X" treatment strategy of early-stage synchronous multiple primary lung cancer

BACKGROUND

Diagnosing and treating synchronous multiple primary lung cancers (sMPLC) are complex and challenging. This study aimed to report real-world data on the comprehensive diagnosis and treatment of patients with early-stage sMPLC.

MATERIALS AND METHODS

A single-center cohort study was carried out and a large number of patients with early-stage sMPLC were included. A single- or two-stage surgery was performed to remove the primary and co-existing lesions. The "X" strategies, including ablation, SBRT, and EGFR-TKIs treatment, were applied to treat the high-risk residual lesions. Wide panel-genomic sequencing was performed to assess the genetic heterogeneity of the co-existing lesions.

RESULTS

A total of 465 early-stage sMPLC patients with 1198 resected lesions were included. Despite most patients being histologically different or harboring different genetic alternations, about 7.5% of the patients had the same histological type and driver gene mutation changes, comprehensive re-evaluation is thus needed. The "Surgery + X" strategy showed remarkable efficacy and safety in treating multiple lesions. Follow-up data revealed that the T2 stage (p = 0.014) and the solid presence of a primary lesion (p < 0.001) were significantly related to tumor recurrence. And a T2-stage primary tumor had a significantly higher rate of developing new lesions after the initial surgery (p < 0.001).

CONCLUSIONS

In real-world practice, histopathological and radiological evaluation combined with genetic analyses could be a robust diagnostic approach for sMPLC. The "Surgery + X" treatment strategy showed remarkable efficacy, superiority, and safety in the clinical treatment of early-stage sMPLC.

Development and validation of a nomogram model for lung cancer based on radiomics artificial intelligence score and clinical blood test data

Background

Artificial intelligence (AI) discrimination models using single radioactive variables in recognition algorithms of lung nodules cannot predict lung cancer accurately. Hence, we developed a clinical model that combines AI with blood test variables to predict lung cancer.

Methods

Between 2018 and 2021, 584 individuals (358 patients with lung cancer and 226 individuals with lung nodules other than cancer as control) were enrolled prospectively. Machine learning algorithms including lasso regression and random forest (RF) were used to select variables from blood test data, Logistic regression analysis was used to reconfirm the features to build the nomogram model. The predictive performance was assessed by performing the receiver operating characteristic (ROC) curve analysis as well as calibration, clinical decision and impact curves. A cohort of 48 patients was used to independently validate the model. The subgroup application was analyzed by pathological diagnosis.

Findings

A total of 584 patients were enrolled (358 lung cancers, 61.30%,226 patients for the control group) to establish the model. The integrated model identified eight potential factors including carcinoembryonic antigen (CEA), AI score, Pro-Gastrin Releasing Peptide (ProGRP), cytokeratin 19 fragment antigen21-1(CYFRA211), squamous cell carcinoma antigen(SCC), indirect bilirubin(IBIL), activated partial thromboplastin time(APTT) and age. The area under the curve (AUC) of the nomogram was 0.907 (95% CI, 0.881-0.929). The decision and clinical impact curves showed good predictive accuracy of the model. An AUC of 0.844 (95% CI, 0.710 - 0.932) was obtained for the external validation group.

Conclusion

The nomogram model integrating AI and clinical data can accurately predict lung cancer, especially for the squamous cell carcinoma subtype.

Preoperative computed tomography-guided localization for multiple lung nodules: a Meta-analysis

PURPOSE

Approximately 20% of patients with lung nodules (LNs) have multiple LNs (MLNs). This meta-analysis was performed to assess the safety and efficacy of computed tomography (CT)-guided localization of MLNs in comparison with those of single LN (SLN) localization.

MATERIAL AND METHODS

The PubMed, Embase, and Cochrane Library were searched to collect relevant articles published till February 2022. The meta-analysis was performed using the RevMan v5.3.

RESULTS

In total, seven studies met the inclusion criteria for this meta-analysis. No significant difference was observed between patients with MLNs and SLN in terms of pooled successful localization rate based on LNs (p = 0.64) and patients (p = 0.06). The pooled duration of localization was significantly shorter and the pooled pneumothorax and lung hemorrhage rates were significantly lower in the SLN group than in the MLNs group (p < 0.00001 for all). The pooled duration of hospital stay was comparable between the MLNs and SLN groups (p = 0.96). Significant heterogeneity was observed in the endpoints of duration of localization (I² = 75%) and pneumothorax (I² = 53%).

CONCLUSIONS

CT-guided simultaneous MLN localization is clinically safe and effective, despite requiring a longer procedural time and having higher incidence of pneumothorax and lung hemorrhage than SLN localization.

Comprehensive analysis of clinical logistic and machine learning based models for the evaluation of pulmonary nodules

Introduction

Over the years, multiple models have been developed for the evaluation of pulmonary nodules (PNs). This study aimed to comprehensively investigate clinical models for estimating the malignancy probability in patients with PNs.

Methods

PubMed, EMBASE, Cochrane Library, and Web of Science were searched for studies reporting mathematical models for PN evaluation until March 2020. Eligible models were summarized, and network meta-analysis was performed on externally validated models (PROSPERO database CRD42020154731). The cut-off value of 40% was used to separate patients into high prevalence (HP) and low prevalence (LP), and a subgroup analysis was performed.

Results

A total of 23 original models were proposed in 42 included articles. Age and nodule size were most often used in the models, whereas results of positron emission tomography-computed tomography were used when collected. The Mayo model was validated in 28 studies. The area under the curve values of four most often used models (PKU, Brock, Mayo, VA) were 0.830, 0.785, 0.743, and 0.750, respectively. High-prevalence group (HP) models had better results in HP patients with a pooled sensitivity and specificity of 0.83 (95% confidence interval [CI]: 0.78–0.88) and 0.71 (95% CI: 0.71–0.79), whereas LP models only achieved pooled sensitivity and specificity of 0.70 (95% CI: 0.60–0.79) and 0.70 (95% CI: 0.62–0.77). For LP patients, the pooled sensitivity and specificity decreased from 0.68 (95% CI: 0.57–0.78) and 0.93 (95% CI: 0.87–0.97) to 0.57 (95% CI: 0.21–0.88) and 0.82 (95% CI: 0.65–0.92) when the model changed from LP to HP models. Compared with the clinical models, artificial intelligence-based models have promising preliminary results.

Conclusions

Mathematical models can facilitate the evaluation of lung nodules. Nevertheless, suitable model should be used on appropriate cohorts to achieve an accurate result.

Artificial intelligence: opportunities in lung cancer

PURPOSE OF REVIEW

In this article, we focus on the role of artificial intelligence in the management of lung cancer. We summarized commonly used algorithms, current applications and challenges of artificial intelligence in lung cancer.

RECENT FINDINGS

Feature engineering for tabular data and computer vision for image data are commonly used algorithms in lung cancer research. Furthermore, the use of artificial intelligence in lung cancer has extended to the entire clinical pathway including screening, diagnosis and treatment. Lung cancer screening mainly focuses on two aspects: identifying high-risk populations and the automatic detection of lung nodules. Artificial intelligence diagnosis of lung cancer covers imaging diagnosis, pathological diagnosis and genetic diagnosis. The artificial intelligence clinical decision-support system is the main application of artificial intelligence in lung cancer treatment. Currently, the challenges of artificial intelligence applications in lung cancer mainly focus on the interpretability of artificial intelligence models and limited annotated datasets; and recent advances in explainable machine learning, transfer learning and federated learning might solve these problems.

SUMMARY

Artificial intelligence shows great potential in many aspects of the management of lung cancer, especially in screening and diagnosis. Future studies on interpretability and privacy are needed for further application of artificial intelligence in lung cancer.