肺腺癌中 TIL（肿瘤浸润淋巴细胞）的工程化特征评估

Tumor-Infiltrating Lymphocytes (TIL) are emerging as immunotherapy prognostic markers. Currently, TIL are assessed on hematoxylin and eosin (H&E)-stained slides of tumor tissue by pathologists. This approach is time-consuming, and subjected to inter-observer variability. The aim of this study is to propose a machine learning-based algorithm, called Feature Engineering TIL Assessment (FTA), for the automatic TIL assessment by using adenocarcinoma metadata (i. e. , anamnestic, clinical and pathological data).

The algorithm is an Elastic Net, tuned by Bayesian Optimization and validated by Leave-One-Subject-Out cross validation. Obtained coefficients were used for feature ranking. Results confirms the goodness of performance of FTA, with an overall Mean Absolute Error of 2. 1%, Concordance Correlation Coefficient equal to 0. 71 and difference in the Bland- Altman plot equal to -0. 001. The obtained feature ranking revealed the key role of gender, as confirmed by the clinical literature.

In conclusion, FTA is the first image-independent automatic TIL assessment procedure, having the potential to address challenges associated with inter-observer variability and the time-consuming nature of classical procedures.