341P Artificial intelligence-supported early detection of lung cancer from chest x-ray in routine clinical practice: Real-world, multicenter study across Czech hospitals
Bibliographic record
Abstract
Background:The increasing complexity of clinical guidelines and physicians' time constraints create an urgent need for robust clinical decision support systems.Much hope rests on the use of large language models (LLMs) but their non-deterministic nature poses significant risks.Combining LLMs with deterministic rule-based algorithms represents a promising alternative.This study compares the performance of a specialised urology LLM with a hybrid LLM-rule-based system for prostate cancer risk stratification and treatment selection. Methods:We developed 142 non-metastatic prostate cancer index cases representing all risk groups.Two senior urooncologists independently classified each case according to the EAU 2025 risk groups and identified therapies with "strong" EAU prostate cancer guideline recommendations.These cases were processed through two distinct systems: the "EAU Guidelines Bot", a specialised ChatGPT-based LLM fine-tuned on the EAU 2025 guidelines, and a self-developed hybrid clinical decision support system employing LLM-based (Deepseek-V3) feature extraction from unstructured text followed by deterministic rule-based risk stratification and therapy recommendations.The output of each system was then compared with the expertdefined ground truth.The evaluation occurred in August 2025. Results:The hybrid LLM-rule based system significantly outperformed the specialised LLM across both evaluation metrics (p < 0.05).For risk group classification, the hybrid system achieved 90.8% accuracy with a macro-averaged F1 score of 0.92, exceeding the EAU Bot's performance of 80.3% accuracy and a macro F1 score of 0.81.Similarly, the hybrid approach demonstrated stronger performance in identifying the correct therapies with 90.8% accuracy and a macro F1 score of 0.92, compared to 78.2% accuracy and a macro F1 score of 0.77 for the EAU Bot. Conclusions:Even specialised LLMs lack sufficient accuracy for autonomous clinical deployment in prostate cancer risk stratification and treatment planning.However, when integrated with rule-based engines, LLMs reliably structure medical data, enabling hybrid systems to achieve exceptional accuracy.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.003 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.001 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".