Integrating Machine Learning for Prognostic Prediction in Stage I NSCLC by CT Images and Pathological Factors

Integrating Machine Learning for Prognostic Prediction in Stage I NSCLC: a Multicenter Analysis

The investigators retrospectively collected the participants with stage I non-small cell lung cancer (NSCLC) patients resected between January 2010 to December 2020 for training and internal validation. The Clinical data, preoperative clinical information, laboratory results and CT images were collected. The investigators also collected the disease-free survival time. On the Deepwise multi-modal research platform, the images were semi-automatically segmented and expanded outward by 3mm to obtain the peritumor tissue. PyRadiomics was used to extract the radiomic features. LASSOcox and rsf were used to select the features. we developed a machine learning-based integrative prognostic model that utilizes radiomic and pathological variables as input using LOOCV framework. And it was further tested on the internal and external cohorts. Discrimination was assessed by using the C-index and area under the receiver operating characteristic curve (AUC), IBS, DCA.

Inclusion Criteria:

patients with stage I NSCLC (ninth AJCC edition) who underwent curative R0 resections between January 2010 and December 2020 -

Exclusion Criteria:

1. absence of enhanced CT
2. history of lung cancer or synchronous lung cancers
3. follow-up records ≤3 Months
4. carcinoma in situ (CIS) or minimally invasive NSCLC
5. death within 30 days of surgery
6. no pathological slides or reports