Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ProKAN: Progressive Stacking of Kolmogorov-Arnold Networks for Efficient Liver Segmentation

Bhavesh Gyanchandani, Aditya Oza, Abhinav Roy

TL;DR

ProKAN addresses the need for accurate and efficient 3D liver segmentation in CT images. It introduces progressive stacking of Kolmogorov-Arnold Networks (KANs) with learnable B-spline activations to adaptively grow model capacity while preventing overfitting. The approach demonstrates state-of-the-art performance on LiTS17 with higher Dice scores and better generalization compared to MLPs and fixed KANs, along with favorable training/inference efficiency. The work suggests practical benefits for real-time clinical deployment and offers interpretable learned spline coefficients.

Abstract

The growing need for accurate and efficient 3D identification of tumors, particularly in liver segmentation, has spurred considerable research into deep learning models. While many existing architectures offer strong performance, they often face challenges such as overfitting and excessive computational costs. An adjustable and flexible architecture that strikes a balance between time efficiency and model complexity remains an unmet requirement. In this paper, we introduce proKAN, a progressive stacking methodology for Kolmogorov-Arnold Networks (KANs) designed to address these challenges. Unlike traditional architectures, proKAN dynamically adjusts its complexity by progressively adding KAN blocks during training, based on overfitting behavior. This approach allows the network to stop growing when overfitting is detected, preventing unnecessary computational overhead while maintaining high accuracy. Additionally, proKAN utilizes KAN's learnable activation functions modeled through B-splines, which provide enhanced flexibility in learning complex relationships in 3D medical data. Our proposed architecture achieves state-of-the-art performance in liver segmentation tasks, outperforming standard Multi-Layer Perceptrons (MLPs) and fixed KAN architectures. The dynamic nature of proKAN ensures efficient training times and high accuracy without the risk of overfitting. Furthermore, proKAN provides better interpretability by allowing insight into the decision-making process through its learnable coefficients. The experimental results demonstrate a significant improvement in accuracy, Dice score, and time efficiency, making proKAN a compelling solution for 3D medical image segmentation tasks.

ProKAN: Progressive Stacking of Kolmogorov-Arnold Networks for Efficient Liver Segmentation

TL;DR

ProKAN addresses the need for accurate and efficient 3D liver segmentation in CT images. It introduces progressive stacking of Kolmogorov-Arnold Networks (KANs) with learnable B-spline activations to adaptively grow model capacity while preventing overfitting. The approach demonstrates state-of-the-art performance on LiTS17 with higher Dice scores and better generalization compared to MLPs and fixed KANs, along with favorable training/inference efficiency. The work suggests practical benefits for real-time clinical deployment and offers interpretable learned spline coefficients.

Abstract

The growing need for accurate and efficient 3D identification of tumors, particularly in liver segmentation, has spurred considerable research into deep learning models. While many existing architectures offer strong performance, they often face challenges such as overfitting and excessive computational costs. An adjustable and flexible architecture that strikes a balance between time efficiency and model complexity remains an unmet requirement. In this paper, we introduce proKAN, a progressive stacking methodology for Kolmogorov-Arnold Networks (KANs) designed to address these challenges. Unlike traditional architectures, proKAN dynamically adjusts its complexity by progressively adding KAN blocks during training, based on overfitting behavior. This approach allows the network to stop growing when overfitting is detected, preventing unnecessary computational overhead while maintaining high accuracy. Additionally, proKAN utilizes KAN's learnable activation functions modeled through B-splines, which provide enhanced flexibility in learning complex relationships in 3D medical data. Our proposed architecture achieves state-of-the-art performance in liver segmentation tasks, outperforming standard Multi-Layer Perceptrons (MLPs) and fixed KAN architectures. The dynamic nature of proKAN ensures efficient training times and high accuracy without the risk of overfitting. Furthermore, proKAN provides better interpretability by allowing insight into the decision-making process through its learnable coefficients. The experimental results demonstrate a significant improvement in accuracy, Dice score, and time efficiency, making proKAN a compelling solution for 3D medical image segmentation tasks.
Paper Structure (19 sections, 6 equations, 3 figures, 7 tables)

This paper contains 19 sections, 6 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methodology
  4. Kolmogorov-Arnold Networks (KANs)
  5. KAN Formulation
  6. Spline Parameterization in KANs
  7. Progressive Stacking in proKAN
  8. Progressive Stacking Mechanism
  9. Overfitting Detection Criteria
  10. Dynamic Hyperparameter Adjustment
  11. Experimental Results
  12. Data
  13. Liver Segmentation Performance Metrics
  14. Accuracy and Overfitting Mitigation
  15. Overfitting Mitigation
  16. ...and 4 more sections

Figures (3)

  • Figure 1: Overall Workflow for Liver Segmentation: The diagram depicts the full pipeline from input CT scan to the final segmented liver tumor output. The process begins with the acquisition of contrast-enhanced abdominal CT scans. These images undergo preprocessing steps such as normalization and noise reduction. The preprocessed images are then passed through the progressive proKAN architecture, which dynamically adjusts its complexity to avoid overfitting. The final output is the segmented liver tumor region.
  • Figure 2: Kolmogorov-Arnold Network (KAN) Architecture: This figure illustrates the architecture of a KAN block used in the proKAN model. Each edge is modeled by a learnable activation function, specifically B-splines, allowing for non-linear transformations. The flexibility of the network increases with deeper layers, enabling complex relationships to be modeled efficiently while maintaining interpretability.
  • Figure 3: Example CT scans from the LiTS17 dataset showcasing liver tumor regions.