Reliable gene mutation prediction in clear cell renal cell carcinoma through multi-classifier multi-objective...

Area of Science:

• Medical Imaging
• Genomics
• Artificial Intelligence

Background:

• Genetic studies link gene mutations to clear cell renal cell carcinoma (ccRCC).
• Non-invasive methods are needed to determine tumor mutation status due to limitations of biopsy and sequencing.
• Radiogenomics, analyzing medical image features, offers a promising alternative for identifying disease genomics.

Purpose of the Study:

• To develop a novel multi-classifier multi-objective (MCMO) radiogenomics model for predicting ccRCC-related gene mutations.
• To improve the reliability of predictive models by simultaneously optimizing for sensitivity and specificity.
• To leverage the strengths of diverse classifiers through an evidential reasoning fusion approach.

Main Methods:

• A multi-classifier multi-objective (MCMO) model was developed using quantitative CT features.
• Similarity-based sensitivity and specificity were used as dual objective functions during training.
• The evidential reasoning (ER) approach fused outputs from multiple classifiers.
• A novel similarity-based multi-objective optimization (SMO) algorithm trained the MCMO model.

Main Results:

• The MCMO model achieved a predictive area under the receiver operating characteristic curve (AUC) exceeding 0.85 for VHL, PBRM1, and BAP1 gene mutations.
• The model demonstrated balanced sensitivity and specificity in predicting these mutations.
• The MCMO approach outperformed individual classifiers and other optimization/fusion strategies.

Conclusions:

• The proposed MCMO radiogenomics model provides a reliable and non-invasive method for predicting key gene mutations in ccRCC.
• This approach enhances diagnostic accuracy by integrating multiple classifiers and objectives.
• Radiogenomics holds significant potential for personalized medicine in ccRCC by bridging imaging and genomic data.