Back

A CT-based radiomics approach for CD8+ lymphocytes infiltrationstratification in patients with non-small cell lung cancer

Zerka, F.; FELFLI, M.; VOYTON, C.; THINNES, A.; JACQUES, S.; LIU, Y.; IANNESSI, A.

2024-05-24 oncology
10.1101/2024.05.23.24307791 medRxiv
Show abstract

BackgroundAccurate prediction of tumor microenvironment is crucial for optimizing decision making throughout cancer treatment process. Current biopsy or surgical-based approaches to assess tumor microenvironment are limited by their invasiveness and tumor heterogeneity. The present study aimed to investigate the association of computed tomography radiomics features and CD8+ lymphocyte infiltration levels for patients with non-small cell lung cancer. Materials and Methods283 patients with CT imaging and RNA-Seq data were collected from open-source data repositories. The study included three independent cohorts of non-small cell lung cancer patients, with one serving as the training set and the other two as external test sets. 1246 CT radiomics features were extracted. Three discriminative texture features were used to train the AI model. ResultsThe model, trained on discriminative features, achieved a mean area under the curve AUC-ROC of 0.71({+/-}0.17 std) on the training data. The AUC-ROC of the model on the two independent test sets is 0.67 (95% CI: 51%, 80%) on TCGA and 0.64 (95% CI: 51%, 74%) on LUNG3. ConclusionCT texture features can differentiate patients with high from low CD8+ lymphocyte infiltration levels. These features can non-invasively analyze the whole tumor and aid in the identification of patients that can respond to immunotherapy. Tweetable abstractTexture radiomics features on CT scans can aid in stratifying CD8+ lymphocyte infiltration levels for patients with NSCLC.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.