Multimodal risk stratification of non-metastatic lung adenocarcinoma using AI on histology and immunohistochemistry slides.
Bibliographic record
Abstract
e20083 Background: While immunohistochemistry (IHC) slides are commonly used to assess the presence of actionable biomarkers, the recent advances in artificial intelligence (AI) focus exclusively on hematoxylin and eosin (H&E) slides. Nonetheless, many IHC slides, potentially rich in prognostic information, remain underexplored for risk stratification and outcome prediction. Our study aims to assess the complementarity of H&E and IHC slides through the development of a multimodal prognostic AI model for lung adenocarcinoma patients. Methods: We collected 1 H&E slide, 8 IHC slides (PD-L1, FOXP3, HELA2, CD3, CD4, CD8, CD163, KI67), and 2 double-stained slides (PDL1-CD163 and CD3-CD163) from 245 resected patients with lung adenocarcinoma at two different hospitals (median OS = 54.0 months [11.3 - 138.0]). Patients underwent various adjuvant treatments (chemotherapy n=92; immunotherapy n=4; other n=36; no treatment n=152). We cross-validated 11 AI models (one model per staining) to predict overall survival in these patients. We used Harell’s C-index and the 70-month cumulative AUC to evaluate the performance of our models. To assess their stratification power, each model’s predictions were split into high-risk and low-risk populations to compute a Hazard Ratio and a log-rank test p-value. Subsequently, a Cox proportional hazards model aggregated the predictions of the 11 models to compute a final risk score. We performed a grid search to find the combination of markers yielding the highest C-index. We calculated the hazard ratios to determine each biomarker's predictive value. Results: The H&E model was the best unimodal model, with a C-index of 0.69 [0.67 - 0.71]. The best-performing multivariate model was obtained by combining the risk scores from univariate models in a Cox regression using H&E, CD8, PD-L1, HELA2, and FOXP3 slides, reaching a C-index of 0.71 [0.69 - 0.73] (Table). We found that the predictions from H&E (HR = 1.39; P = 0.03) and PD-L1 (HR = 1.17; P = 0.04) were significantly associated with survival. Conclusions: By significantly improving patient stratification with non-metastatic lung adenocarcinoma, our multivariate model proves that some IHC markers contain complementary proteomic information to H&E slides. Consequently, this study paves the way for developing more comprehensive risk assessment tools from H&E and IHC stainings. [Table: see text]
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".