Multi-group diagnostic classification of high-dimensional data using differential scanning calorimetry plasma...

Area of Science:

• Biophysics
• Medical Diagnostics
• Statistical Modeling

Background:

• Differential scanning calorimetry (DSC) measures protein denaturation in blood plasma.
• Analyzing high-dimensional DSC thermograms for health status classification is challenging.
• Previous studies have not fully explored high-dimensional data for cancer classification.

Purpose of the Study:

• To develop a statistical data reduction approach and a parametric method (PM) for classifying high-dimensional DSC data.
• To compare the PM's performance against K-nearest neighbors (KNN) and KNN with dynamic time warping (DTW) for cancer detection.
• To evaluate classification accuracy using DSC and demographic data for two- and three-group cancer classifications.

Main Methods:

• Applied DSC to blood plasma samples to obtain protein denaturation thermograms.
• Developed a statistical approach for data reduction and a parametric method (PM) for classification.
• Compared PM with KNN and DTW-KNN for classifying normal vs. cancer (cervical, lung) and normal vs. cervical vs. lung cancer.

Main Results:

• Two-group cancer classifications achieved high performance; three-group classification was more challenging.
• All methods identified normal samples more accurately than cancer samples.
• PM demonstrated higher or equal specificity than KNN/DTW-KNN, with lower sensitivity. PM performance was robust to demographic data inclusion, unlike KNN/DTW-KNN.

Conclusions:

• The proposed statistical approach and PM are effective for high-dimensional DSC data analysis in cancer classification.
• PM offers advantages over KNN and DTW-KNN, particularly in reduced susceptibility to overfitting and improved accuracy with more data.
• Further improvements in PM accuracy are achievable by increasing data points and incorporating additional clinical data.