Optimizing the timing of diagnostic testing after positive findings in lung cancer screening: a proof of concept radiomics study

BACKGROUND

The timeliness of diagnostic testing after positive screening remains suboptimal because of limited evidence and methodology, leading to delayed diagnosis of lung cancer and over-examination. We propose a radiomics approach to assist with planning of the diagnostic testing interval in lung cancer screening.

METHODS

From an institute-based lung cancer screening cohort, we retrospectively selected 92 patients with pulmonary nodules with diameters ≥ 3 mm at baseline (61 confirmed as lung cancer by histopathology; 31 confirmed cancer-free). Four groups of region-of-interest-based radiomic features (n = 310) were extracted for quantitative characterization of the nodules, and eight features were proven to be predictive of cancer diagnosis, noise-robust, phenotype-related, and non-redundant. A radiomics biomarker was then built with the random survival forest method. The patients with nodules were divided into low-, middle- and high-risk subgroups by two biomarker cutoffs that optimized time-dependent sensitivity and specificity for decisions about diagnostic workup within 3 months and about repeat screening after 12 months, respectively. A radiomics-based follow-up schedule was then proposed. Its performance was visually assessed with a time-to-diagnosis plot and benchmarked against lung RADS and four other guideline protocols.

RESULTS

The radiomics biomarker had a high time-dependent area under the curve value (95% CI) for predicting lung cancer diagnosis within 12 months; training: 0.928 (0.844, 0.972), test: 0.888 (0.766, 0.975); the performance was robust in extensive cross-validations. The time-to-diagnosis distributions differed significantly between the three patient subgroups, p < 0.001: 96.2% of high-risk patients (n = 26) were diagnosed within 10 months after baseline screen, whereas 95.8% of low-risk patients (n = 24) remained cancer-free by the end of the study. Compared with the five existing protocols, the proposed follow-up schedule performed best at securing timely lung cancer diagnosis (delayed diagnosis rate: < 5%) and at sparing patients with cancer-free nodules from unnecessary repeat screenings and examinations (false recommendation rate: 0%).

CONCLUSIONS

Timely management of screening-detected pulmonary nodules can be substantially improved with a radiomics approach. This proof-of-concept study's results should be further validated in large programs.

Radiomics-based machine learning model for efficiently classifying transcriptome subtypes in glioblastoma patients from MRI

BACKGROUND

In the field of glioma, transcriptome subtypes have been considered as an important diagnostic and prognostic biomarker that may help improve the treatment efficacy. However, existing identification methods of transcriptome subtypes are limited due to the relatively long detection period, the unattainability of tumor specimens via biopsy or surgery, and the fleeting nature of intralesional heterogeneity. In search of a superior model over previous ones, this study evaluated the efficiency of eXtreme Gradient Boosting (XGBoost)-based radiomics model to classify transcriptome subtypes in glioblastoma patients.

METHODS

This retrospective study retrieved patients from TCGA-GBM and IvyGAP cohorts with pathologically diagnosed glioblastoma, and separated them into different transcriptome subtypes groups. GBM patients were then segmented into three different regions of MRI: enhancement of the tumor core (ET), non-enhancing portion of the tumor core (NET), and peritumoral edema (ED). We subsequently used handcrafted radiomics features (n = 704) from multimodality MRI and two-level feature selection techniques (Spearman correlation and F-score tests) in order to find the features that could be relevant.

RESULTS

After the feature selection approach, we identified 13 radiomics features that were the most meaningful ones that can be used to reach the optimal results. With these features, our XGBoost model reached the predictive accuracies of 70.9%, 73.3%, 88.4%, and 88.4% for classical, mesenchymal, neural, and proneural subtypes, respectively. Our model performance has been improved in comparison with the other models as well as previous works on the same dataset.

CONCLUSION

The use of XGBoost and two-level feature selection analysis (Spearman correlation and F-score) could be expected as a potential combination for classifying transcriptome subtypes with high performance and might raise public attention for further research on radiomics-based GBM models.