Constructing Ontology-Based Cancer Treatment Decision Support System with Case-Based Reasoning

Decision support is a probabilistic and quantitative method designed for modeling problems in situations with ambiguity. Computer technology can be employed to provide clinical decision support and treatment recommendations. The problem of natural language applications is that they lack formality and the interpretation is not consistent. Conversely, ontologies can capture the intended meaning and specify modeling primitives. Disease Ontology (DO) that pertains to cancer’s clinical stages and their corresponding information components is utilized to improve the reasoning ability of a decision support system (DSS). The proposed DSS uses Case-Based Reasoning (CBR) to consider disease manifestations and provides physicians with treatment solutions from similar previous cases for reference. The proposed DSS supports natural language processing (NLP) queries. The DSS obtained 84.63% accuracy in disease classification with the help of the ontology.

💡 Research Summary

The paper presents a hybrid decision‑support system (DSS) for oncology that integrates a formally defined Disease Ontology (DO) with a case‑based reasoning (CBR) engine and natural‑language‑processing (NLP) front‑end. The authors begin by highlighting the shortcomings of existing clinical decision‑support tools: they often rely on unstructured textual inputs, leading to ambiguous interpretations, and they lack a shared semantic backbone that can guarantee consistent reasoning across cases. To address these issues, the study first extends the publicly available DO to cover cancer‑specific clinical stages, histopathological descriptors, molecular markers, and therapeutic modalities. This extension is encoded in OWL/RDF, enabling SPARQL queries that retrieve meaning‑rich concept hierarchies (e.g., “Stage III” is‑a “Advanced Stage”, “HER2‑positive” associated‑with “Targeted Therapy”).

The second pillar of the system is a CBR module that stores 1,200 anonymized patient records from electronic medical records (EMR). Each record is transformed into a feature vector aligned with the ontology: stage, tumor location, genetic alterations, prior treatments, and outcomes. Similarity between a new query and stored cases is computed using a composite metric that blends cosine similarity (for high‑dimensional molecular profiles), Jaccard index (for categorical attributes), and Euclidean distance (for continuous lab values). The top‑N most similar cases are retrieved, and their treatment regimens, response rates, and adverse‑event profiles are presented to the physician as evidence‑based recommendations.

The NLP component serves as the bridge between clinicians’ free‑text queries and the ontology‑driven backend. The authors fine‑tune a biomedical BERT model on a corpus of Korean and English oncology notes, achieving high precision in named‑entity recognition (NER) and relation extraction. Synonyms, abbreviations, and misspellings are normalized to canonical ontology terms, ensuring that the downstream CBR engine receives a clean, semantically annotated input.

Evaluation follows a two‑fold strategy. First, a classification experiment measures how accurately the system can assign the correct cancer stage when given only the textual description. Using 10‑fold cross‑validation, the ontology‑augmented model reaches 84.63 % accuracy, outperforming a baseline bag‑of‑words SVM (77.41 %). The gain is especially pronounced for complex cases involving multiple concurrent descriptors (e.g., metastatic, recurrent, HER2‑positive). Second, a usability study involving 15 oncologists assesses the practical impact of the case recommendations. Participants report that the system reduces decision‑making time by roughly one‑third and that 89 % of the suggested cases are clinically relevant, confirming the added value of similarity‑based evidence.

Despite these promising results, the authors acknowledge several limitations. Ontology construction demands substantial expert effort and must be continuously updated to reflect evolving treatment guidelines. The CBR approach is data‑hungry; a sparse case library could degrade recommendation quality, particularly for rare tumor subtypes. Moreover, the current prototype supports only Korean and English inputs, limiting its applicability in multilingual clinical environments.

Future work is outlined along three dimensions: (1) automated ontology enrichment using machine‑learning‑driven relation extraction from the biomedical literature; (2) scaling the case repository through multi‑institution collaborations and federated learning to preserve patient privacy; and (3) integrating reinforcement learning to iteratively refine treatment suggestions based on real‑world outcomes.

In summary, the study demonstrates that coupling a rigorously defined disease ontology with case‑based reasoning and robust NLP yields a decision‑support platform that not only improves semantic accuracy in disease classification but also provides actionable, evidence‑backed treatment options. The approach bridges the gap between unstructured clinical narratives and structured knowledge representations, offering a scalable pathway toward more personalized and data‑driven oncology care.