Fig. 4
Comprehensive methodology pipeline for choroidal melanoma detection using tear fluid mass spectrometry data. The workflow consists of seven sequential stages: (1) Data Collection - Non-invasive tear fluid sampling using Schirmer strips from 6 choroidal melanoma patients and 6 healthy controls; (2) Protein Corona Formation - Synthesis of 25 nm gold nanoparticles via Turkevich method and protein corona formation at 37 °C; (3) Mass Spectrometry Analysis - ESI-MS detection of protein fragments after trypsin digestion; (4) Data Augmentation - Three-fold dataset expansion using Gaussian noise, scaling, and random shifts; (5) Signal Preprocessing - Segmentation into 128-point windows and extraction of 8 statistical and entropy-based features; (6) Parallel Model Training - Branch A: Traditional machine learning algorithms (SVM, RF, DT, DNN) using extracted features; Branch B: Transfer learning with pretrained CNNs (VGG16, ResNet50, Xception) using CWT-transformed 128 × 128 RGB images; (7) Performance Evaluation − 5-fold cross-validation assessment using accuracy, sensitivity, precision, F1-score, AUC-ROC, and specificity metrics.
