Knowledge-Infused Abstractive Summarization of Clinical Diagnostic Interviews: Framework Development Study

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Clinical Psychology

Background:

• Clinical diagnostic interviews are complex, with mental health professionals (MHPs) gathering crucial data amidst patient discomfort and social stigma.
• Effective summarization of these interviews is vital for MHPs to make informed decisions and explore patient behavior deeply, especially in critical situations.

Purpose of the Study:

• To propose an unsupervised, knowledge-infused abstractive summarization (KiAS) approach for clinical interviews.
• To enhance existing summarization methods by generating more informative summaries than those based on frequency heuristics.
• To enable MHPs to conduct better-informed follow-ups with patients.

Main Methods:

• Incorporated domain knowledge from the Patient Health Questionnaire-9 lexicon into an integer linear programming framework.
• Optimized for linguistic quality and informativeness in summary generation.
• Evaluated KiAS against three baseline summarization approaches (SumBasic, abstractive ILP, abstraction over extractive) using the Distress Analysis Interview Corpus-Wizard of Oz dataset.

Main Results:

• KiAS generated concise (7 sentences) yet informative summaries from long, ambiguous clinical interviews (58 sentences).
• Achieved significant improvements over baselines in thematic overlap (23.3%), Flesch Reading Ease (4.4%), contextual similarity (2.5%), and Jensen Shannon divergence (2.2%).
• Demonstrated substantial gains in ROUGE-2 (61%) and ROUGE-L (49%) metrics, validated by mental health professionals.

Conclusions:

• KiAS shows potential utility in leveraging voluminous patient communications outside scheduled appointments.
• The approach promises to generate semantically relevant summaries, aiding MHPs in making informed decisions about patient status.
• This method can significantly impact clinical practice by improving the efficiency and depth of patient information analysis.