Abstract 463: Highly accurate machine learning assessment of immune-related pathologic response criteria (irPRC) scoring in patients with non-small cell lung carcinoma (NSCLC) treated with neoadjuvant anti-PD-1-based therapies
Bibliographic record
Abstract
Abstract Pathological complete response (no residual viable tumor, RVT) and/or major pathologic response (≤10% RVT) are now primary or secondary endpoints for a large proportion of clinical trials studying neoadjuvant immunotherapeutic regimens. We previously developed a scoring system for assessing pathologic response after immunotherapy, termed irPRC (Cottrell et al. Ann Oncol 2018). By these criteria, %RVT is assessed by dividing RVT by the sum of the surface area on the slide composed of RVT + necrosis + regression bed– the latter feature is where the tumor used to be and is characterized by fibroinflammatory stroma that is distinct from tumoral stroma. We have previously reported high inter-observer reproducibility for pathologic response assessment following immunotherapy. However, these assessments involve performing evaluations that are currently outside the scope of routine surgical pathology training and may be time-consuming. To date, these assessments have primarily been performed by academic pathologists who have seen the largest number of these cases as a part of clinical trials. A machine learning (ML)-powered assessment of irPRC would allow for faster, standardized evaluation and expanded access to patients treated outside of large academic centers. We trained a supervised convolutional neural network to assess pathologic response using irPRC on n=92 H&E-stained slides from patients with advanced, resectable NSCLC treated with neoadjuvant anti-PD-1 +/- anti-CTLA-4 at a single institution. The ML algorithm was trained based on ground-truth manual annotations by pathologists on whole slide digital scans and tested using leave-one-out cross validation. Each of ~830,000 image tiles was classified into one of four classes: tumor, necrosis, immune-mediated regression, or background lung tissue. Receiver operating curves showed that the algorithm exhibited high accuracy for predicting the various tissue classes with an area under the curve of 0.95, 0.96, 0.90, and 0.90 for the four classes, respectively. %RVT was calculated by dividing the surface area of RVT by total tumor bed surface area (RVT + necrosis + regression). There was a strong positive correlation between the machine assessed RVT and the human assessed RVT at both the slide level and case level (aggregate %RVT based on surface area from all slides for a given patient), Pearson’s r=0.95 and r=0.99, respectively. Here, we demonstrate that a ML algorithm performs as well as an experienced pathologist assessment in scoring pathologic response. These findings will need to be validated in larger studies. Additionally, the association of pathologic response with longer term patient outcomes will be evaluated as survival data matures to determine whether pathologic response is a robust surrogate of survival. Citation Format: Julie E. Stein, Vinay Pulim, Tricia R. Cottrell, Patrick M. Forde, Janis M. Taube. Highly accurate machine learning assessment of immune-related pathologic response criteria (irPRC) scoring in patients with non-small cell lung carcinoma (NSCLC) treated with neoadjuvant anti-PD-1-based therapies [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 463.
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 | 0.002 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.000 |
| Open science | 0.000 | 0.000 |
| 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".