KEY TAKEAWAYS
- The study aimed to create and validate ML models diagnosing cervical LNM in NSCLC using ultrasound radiomics and semantic features.
- The results indicated that ultrasound radiomic models, integrated with semantic features, outperform LR in diagnosing cervical LNM in NSCLC.
The prognosis for individuals with non-small cell lung cancer (NSCLC) worsens when cancer spreads to the lymph nodes (LN) in their neck, known as cervical lymph node metastasis (LNM). Accurate identification of LNM in these patients is crucial.
Zhiqiang Deng and the team aimed to create and validate machine learning models utilizing ultrasound radiomic data and descriptive semantic features for diagnosing cervical LNM in patients with NSCLC.
The study enrolled patients with NSCLC who received neck ultrasound examinations followed by cervical LN biopsies across three institutes from January 2019 to January 2022. Radiomic features were extracted from ultrasound images at the maximum cross-sectional areas of cervical LNs. Logistic regression (LR) and random forest (RF) models were constructed. Model efficacy was evaluated using the area under the curve (AUC) and accuracy, internally validated via fivefold cross-validation, and externally validated through a hold-out method.
Of 313 patients with a median age of 64, 276 (88.18%) had cervical LNM. Multivariate analysis identified three descriptive semantic features which were long diameter, shape, and corticomedullary boundary. Among 474 radiomic features, 9 were deemed suitable for the LR model and 15 for the RF model. The average AUCs for semantic and radiomic models were 0.876 (range: 0.781-0.961) and 0.883 (range: 0.798-0.966), respectively.
However, the combined LR model incorporating semantic-radiomics yielded a higher average AUC of 0.901 (range: 0.862-0.927). Utilizing the RF algorithm improved the average AUCs to 0.908 (range: 0.837-0.966) for radiomics and 0.922 (range: 0.872-0.982) for the combined semantic-radiomics model. Validation through the hold-out method produced similar outcomes, with the semantic-radiomics combined RF model achieving the highest AUC value of 0.901 (95% CI, 0.886-0.968).
The study concluded that integrating descriptive semantic features with ultrasound radiomic models holds promise for accurately diagnosing cervical LNM in patients with NSCLC. Additionally, the RF model exhibited superior performance compared to the conventional LR model in diagnosing cervical LNM in this population.
The study received funding from the College Students’ Innovative Entrepreneurial Training Plan Program in Sichuan Province.
Source: https://pubmed.ncbi.nlm.nih.gov/38678211/
Deng Z, Liu X, Wu R, et al. (2024) “Ultrasound-based radiomics machine learning models for diagnosing cervical lymph node metastasis in patients with non-small cell lung cancer: a multicentre study.” BMC Cancer. 2024 Apr 27;24(1):536. doi: 10.1186/s12885-024-12306-6. PMID: 38678211.
