Vít Nováček

Adrianna Janik, Maria Torrente, Luca Costabello et al.

Background: Stratifying cancer patients according to risk of relapse can personalize their care. In this work, we provide an answer to the following research question: How to utilize machine learning to estimate probability of relapse in early-stage non-small-cell lung cancer patients? Methods: For predicting relapse in 1,387 early-stage (I-II), non-small-cell lung cancer (NSCLC) patients from the Spanish Lung Cancer Group data (65.7 average age, 24.8% females, 75.2% males) we train tabular and graph machine learning models. We generate automatic explanations for the predictions of such models. For models trained on tabular data, we adopt SHAP local explanations to gauge how each patient feature contributes to the predicted outcome. We explain graph machine learning predictions with an example-based method that highlights influential past patients. Results: Machine learning models trained on tabular data exhibit a 76% accuracy for the Random Forest model at predicting relapse evaluated with a 10-fold cross-validation (model was trained 10 times with different independent sets of patients in test, train and validation sets, the reported metrics are averaged over these 10 test sets). Graph machine learning reaches 68% accuracy over a 200-patient, held-out test set, calibrated on a held-out set of 100 patients. Conclusions: Our results show that machine learning models trained on tabular and graph data can enable objective, personalised and reproducible prediction of relapse and therefore, disease outcome in patients with early-stage NSCLC. With further prospective and multisite validation, and additional radiological and molecular data, this prognostic model could potentially serve as a predictive decision support tool for deciding the use of adjuvant treatments in early-stage lung cancer. Keywords: Non-Small-Cell Lung Cancer, Tumor Recurrence Prediction, Machine Learning

Douwe J. Spaanderman, Karthik Prathaban, Petr Zelina et al.

Large language models (LLMs) are increasingly used to extract structured information from free-text clinical records, but prior work often focuses on single tasks, limited models, and English-language reports. We evaluated 15 open-weight LLMs on pathology and radiology reports across six use cases, colorectal liver metastases, liver tumours, neurodegenerative diseases, soft-tissue tumours, melanomas, and sarcomas, at three institutes in the Netherlands, UK, and Czech Republic. Models included general-purpose and medical-specialised LLMs of various sizes, and six prompting strategies were compared: zero-shot, one-shot, few-shot, chain-of-thought, self-consistency, and prompt graph. Performance was assessed using task-appropriate metrics, with consensus rank aggregation and linear mixed-effects models quantifying variance. Top-ranked models achieved macro-average scores close to inter-rater agreement across tasks. Small-to-medium general-purpose models performed comparably to large models, while tiny and specialised models performed worse. Prompt graph and few-shot prompting improved performance by ~13%. Task-specific factors, including variable complexity and annotation variability, influenced results more than model size or prompting strategy. These findings show that open-weight LLMs can extract structured data from clinical reports across diseases, languages, and institutions, offering a scalable approach for clinical data curation.

Petr Zelina, Jana Halámková, Vít Nováček

This work is motivated by the scarcity of tools for accurate, unsupervised information extraction from unstructured clinical notes in computationally underrepresented languages, such as Czech. We introduce a stepping stone to a broad array of downstream tasks such as summarisation or integration of individual patient records, extraction of structured information for national cancer registry reporting or building of semi-structured semantic patient representations for computing patient embeddings. More specifically, we present a method for unsupervised extraction of semantically-labelled textual segments from clinical notes and test it out on a dataset of Czech breast cancer patients, provided by Masaryk Memorial Cancer Institute (the largest Czech hospital specialising in oncology). Our goal was to extract, classify (i.e. label) and cluster segments of the free-text notes that correspond to specific clinical features (e.g., family background, comorbidities or toxicities). The presented results demonstrate the practical relevance of the proposed approach for building more sophisticated extraction and analytical pipelines deployed on Czech clinical notes.

Vit Novacek

We introduce an approach to discovery informatics that uses so called knowledge graphs as the essential representation structure. Knowledge graph is an umbrella term that subsumes various approaches to tractable representation of large volumes of loosely structured knowledge in a graph form. It has been used primarily in the Web and Linked Open Data contexts, but is applicable to any other area dealing with knowledge representation. In the perspective of our approach motivated by the challenges of discovery informatics, knowledge graphs correspond to hypotheses. We present a framework for formalising so called hypothesis virtues within knowledge graphs. The framework is based on a classic work in philosophy of science, and naturally progresses from mostly informative foundational notions to actionable specifications of measures corresponding to particular virtues. These measures can consequently be used to determine refined sub-sets of knowledge graphs that have large relative potential for making discoveries. We validate the proposed framework by experiments in literature-based discovery. The experiments have demonstrated the utility of our work and its superiority w.r.t. related approaches.

Vit Novacek

CoCoE stands for Complexity, Coherence and Entropy, and presents an extensible methodology for empirical analysis of Linked Open Data (i.e., RDF graphs). CoCoE can offer answers to questions like: Is dataset A better than B for knowledge discovery since it is more complex and informative?, Is dataset X better than Y for simple value lookups due its flatter structure?, etc. In order to address such questions, we introduce a set of well-founded measures based on complementary notions from distributional semantics, network analysis and information theory. These measures are part of a specific implementation of the CoCoE methodology that is available for download. Last but not least, we illustrate CoCoE by its application to selected biomedical RDF datasets.

Vit Novacek

The paper introduces a framework for representation and acquisition of knowledge emerging from large samples of textual data. We utilise a tensor-based, distributional representation of simple statements extracted from text, and show how one can use the representation to infer emergent knowledge patterns from the textual data in an unsupervised manner. Examples of the patterns we investigate in the paper are implicit term relationships or conjunctive IF-THEN rules. To evaluate the practical relevance of our approach, we apply it to annotation of life science articles with terms from MeSH (a controlled biomedical vocabulary and thesaurus).