Machine learning is an important artificial intelligence technique that is widely applied in cancer diagnosis and detection. More recently, with the rise of personalised and precision medicine, there is a growing trend towards machine learning applications for prognosis prediction. However, to date, building reliable prediction models of cancer outcomes in everyday clinical practice is still a hurdle. In this work, we integrate genomic, clinical and demographic data of lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) patients from The Cancer Genome Atlas (TCGA) and introduce copy number variation (CNV) and mutation information of 15 selected genes to generate predictive models for recurrence and survivability. We compare the accuracy and benefits of three well-established machine learning algorithms: decision tree methods, neural networks and support vector machines. Although the accuracy of predictive models using the decision tree method has no significant advantage, the tree models reveal the most important predictors among genomic information (e.g. KRAS, EGFR, TP53), clinical status (e.g. TNM stage and radiotherapy) and demographics (e.g. age and gender) and how they influence the prediction of recurrence and survivability for both early stage LUAD and LUSC. The machine learning models have the potential to help clinicians to make personalised decisions on aspects such as follow-up timeline and to assist with personalised planning of future social care needs.

The diagnostic performance difference between digital breast tomosynthesis (DBT) and conventional full-field digital mammography (FFDM) for breast suspicious calcifications from various populations is unclear. The objective of this study is to determine whether DBT exhibits the diagnostic advantage for breast suspicious calcifications from various populations compared with FFDM. Three hundred and five patients were enrolled (of which seven patients with bilateral lesions) and 312 breasts images were retrospectively analyzed by three radiologists independently. The postoperative pathology of breast calcifications was the gold standard. Breast cancer was diagnosed utilizing DBT and FFDM with sensitivities of 92.9% and 88.8%, specificities of 87.9% and 75.2%, positive predictive values of 77.8% and 62.1%, negative predictive values of 96.4% and 93.6%, respectively. DBT exhibited significantly higher diagnostic accuracy for benign calcifications compared with FFDM (87.9% vs 75.2%), and no advantage in the diagnosis of malignant calcifications. DBT diagnostic accuracy was notably higher than FFDM in premenopausal (88.4% vs 78.8%), postmenopausal (90.2% vs 77.2%), and dense breast cases (89.4% vs 81.9%). There was no significant difference in non-dense breast cases. In our study, DBT exhibited a superior advantage in dense breasts and benign calcifications cases compared to FFDM, while no advantage was observed in non-dense breasts or malignant calcifications cases. Thus, in the breast cancer screening for young women with dense breasts, DBT may be recommended for accurate diagnosis. Our findings may assist the clinicians in applying the optimal techniques for different patients and provide a theoretical basis for the update of breast cancer screening guideline.

Circulating microRNAs (miRNAs) are promising biomarkers for cancer detection. However, multiethnic and multicentric studies of non-small-cell lung cancer (NSCLC) are lacking. We recruited 221 NSCLC patients, 161 controls and 56 benign nodules from both China and America. Initial miRNA screening was performed using the TaqMan Low Density Array followed by confirming individually by RT-qPCR in Chinese cohorts. Finally, we performed a blind trial from an American cohort to validate our findings. RT-qPCR confirmed that miR-483-5p, miR-193a-3p, miR-25, miR-214 and miR-7 were significantly elevated in patients compared to controls. The areas under the curve (AUCs) of the ROC curve of this five-serum miRNA panel were 0.976 (95% CI, 0.939-1.0; P < 0.0001) and 0.823 (95% CI, 0.75-0.896; P < 0.0001) for the two confirmation sets, respectively. In the blind trial, the panel correctly classified 95% NSCLC cases and 84% controls from the American cohort. Most importantly, the panel was capable of distinguishing NSCLC from benign nodules with an AUC of 0.979 (95% CI, 0.959-1.0) in the American cohort and allowed correct prediction of 86% and 95% stage I-II tumors in the Chinese and American cohorts, respectively. This serum miRNA panel holds the potential for diagnosing ethnically diverse NSCLC patients.