Research on Cancer Prediction Based on Feature Optimization and Multimodal Fusion

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Current lung cancer diagnosis faces challenges due to uneven medical resources, delayed diagnosis, and reliance on single data types.
• Existing screening methods struggle with large-scale application, multitracking, and integrating complementary information.
• Technical hurdles in data collection, multimodal fusion, and AI model development limit clinical use.

Purpose of the Study:

• To develop an efficient, fast, accurate, and radiation-free preliminary lung cancer diagnostic method.
• To explore the integration of AI, novel sensors, and existing data for improved early diagnosis.
• To overcome limitations of current diagnostic approaches by leveraging multisource complementary information.

Main Methods:

• Collected hematological data and utilized fiber-optic vibration and audio sensors to capture lung respiration signals.
• Extracted and optimized features including respiratory frequency, audio-respiratory rhythm, and leukocyte-related hematological markers.
• Employed a SCCA-LMF fusion method for multimodal data integration, feeding into an improved stacking ensemble learning model for binary classification.

Main Results:

• Achieved high predictive accuracy (97.70%), sensitivity (95.75%), specificity (99.64%), and F1 score (99.64%) in a study of 360 participants.
• The developed multimodal model outperformed existing independent methods.
• Demonstrated effective integration of respiratory vibration, audio signals, and routine blood tests, with a multimodal feature grading fusion strategy for 3D data analysis.

Conclusions:

• The study presents a promising AI-driven method for preliminary lung cancer identification.
• This approach effectively bridges engineering and medical fields to enhance healthcare outcomes.
• The reproducible results highlight the potential for improving timely treatment and patient prognosis.