Comprehensive Model Evaluation
Systematic testing of 70 model combinations across two feature sets
5 Classifiers
  • Logistic Regression
  • Naïve Bayes
  • Support Vector Machine
  • Decision Tree
  • Random Forest
×
7 Methods
  • LASSO
  • SelectKBest
  • Particle Swarm Optimization (PSO)
  • Whale Optimization Algorithm (WOA)
  • Grey Wolf Optimizer (GWO)
  • Genetic Algorithm (GA)
  • Simulated Annealing (SA)
=
35 Per Phase
↓
PHASE 1
Radiomics Only (103 features)
35
combinations tested
PHASE 2
Radiomics + Clinical (111 features)
35
combinations tested
↓
COMPREHENSIVE EVALUATION
70
Model Combinations Systematically Tested
Optimal Model Selection
Best performing combination from 70 systematic evaluations

WINNING MODEL

Logistic Regression + Grey Wolf Optimizer

Using Radiomics + Clinical Features (111 features → 10 selected)

Test AUC
0.81
[0.62 - 0.95]
Train AUC
0.79
[0.71 - 0.86]
Selected Features
10
Sparse signature
The Sparse 10-Feature Signature
4
CLINICAL
• Age
• AJCC Stage
• T Stage
• Location
+
6
RADIOMIC
SHAPE (2):
• Maximum2DDiameterSlice
• MinorAxisLength
TEXTURE (4):
• LargeDependenceEmphasis
• RunLengthNonUniformityNorm
• Idm
• Imc1
=
10
FEATURES
Sparse Signature