DEBI-NN: Genetic algorithm vs. gradient descent data

In this study, we compared genetic algorithm (GA) and gradient descent (GD) for training the DEBI-NN architecture, a neural network model based on the Euclidian spatial encoding of neurons in order to drastically reduce the number of trainable parameters. Across all evaluated datasets (synthetic and clinical), GA consistently outperformed GD in terms of balanced accuracy, sensitivity, and specificity, both approaches being optimized in terms of hyperparameters. Specifically, GA achieved 100% balanced accuracy on the synthetic task (vs 83% for GD) and consistently higher performance across the clinical datasets (DLBCL: 83% vs 78%; HECKTOR: 80% vs 67%; Fetal: 81% vs 66%). These findings confirm the limitations of gradient-based optimization for architectures with highly interdependent parameters, as is the case for DEBI-NN, where modifying a single neuron’s position affects all its associated weights. GA thus appears better suited to navigating the resulting non-convex landscape by leveraging its population-based exploratory dynamics.
Four datasets were relied on for the experimental framework established to compare the performance of GA and GD on the DEBI-NN architecture.

- Synthetic Dataset (1000 samples): A simple, non-linear 2D dataset designed to assess the capacity of both optimization algorithms to learn non-linear decision boundaries.
- HECKTOR 2022 Dataset (HPV): This dataset was collected and curated within the context of the 2022 MICCAI challenge (HECKTOR) [1]. It contains 232 18F-FluoroDeoxyGlucose (FDG) Positron Emission Tomography/Computed Tomography (PET/CT) images. The input data combined 8 clinical variables (age, gender, etc.) and 28 Image Biomarker Standardization Initiative (IBSI)-compliant 3D radiomic features extracted from the delineated tumor volumes in both the FDG PET and CT scans.
- DLBCL Dataset (Diffuse Large B-Cell Lymphoma) [2]: 85 cases with both clinical data (7 features) and PET/CT scans from which 10 radiomic features were extracted. The goal was to predict 2-year event-free survival (EFS) as a binary task.
- Fetal Cardiotocography Dataset [3]: 2,126 fetal cardiotocograms (CTGs) characterized by 21 features, with labels obtained through a consensus of three expert obstetricians. The task here consists in classifying the fetal state as Normal, Suspect, or Pathologic.

An executable is available to reproduce the experiments.

[1] Andrearczyk, Vincent, et al. "Overview of the HECKTOR challenge at MICCAI 2022: automatic head and neck tumor segmentation and outcome prediction in PET/CT." 3D Head and Neck Tumor Segmentation in PET/CT Challenge. Cham: Springer Nature Switzerland, 2022. 1-30.

[2] Ritter, Zsombor, et al. "Two-year event-free survival prediction in DLBCL patients based on in vivo radiomics and clinical parameters." Frontiers in Oncology 12 (2022): 820136.

[3] Campos, D. and J. Bernardes. "Cardiotocography." UCI Machine Learning Repository, 2000, https://doi.org/10.24432/C51S4N.