Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NYCTALE: Neuro-Evidence Transformer for Adaptive and Personalized Lung Nodule Invasiveness Prediction

Sadaf Khademi, Anastasia Oikonomou, Konstantinos N. Plataniotis, Arash Mohammadi

TL;DR

NYCTALE introduces a neuro-inspired evidence-accumulation Transformer for adaptive, slice-by-slice prediction of NSCLC invasiveness from CT sequences. The framework comprises a Shifting Window Transformer-based Evidence Encoder and a Drift-Diffusion Model–like Evidence Accumulation Module, outputting a decision only after accumulating sufficient evidence. On an in-house dataset of 114 SSNs, NYCTALE demonstrates competitive accuracy (~81–82%) and AUC (~0.77) while using substantially fewer slices than conventional pooling-based methods, highlighting data efficiency and potential interpretability advantages. By integrating cognitive neuroscience principles with transformer architectures, the approach advances personalized, efficient, and potentially more explainable lung cancer diagnosis within the PM paradigm.

Abstract

Drawing inspiration from the primate brain's intriguing evidence accumulation process, and guided by models from cognitive psychology and neuroscience, the paper introduces the NYCTALE framework, a neuro-inspired and evidence accumulation-based Transformer architecture. The proposed neuro-inspired NYCTALE offers a novel pathway in the domain of Personalized Medicine (PM) for lung cancer diagnosis. In nature, Nyctales are small owls known for their nocturnal behavior, hunting primarily during the darkness of night. The NYCTALE operates in a similarly vigilant manner, i.e., processing data in an evidence-based fashion and making predictions dynamically/adaptively. Distinct from conventional Computed Tomography (CT)-based Deep Learning (DL) models, the NYCTALE performs predictions only when sufficient amount of evidence is accumulated. In other words, instead of processing all or a pre-defined subset of CT slices, for each person, slices are provided one at a time. The NYCTALE framework then computes an evidence vector associated with contribution of each new CT image. A decision is made once the total accumulated evidence surpasses a specific threshold. Preliminary experimental analyses conducted using a challenging in-house dataset comprising 114 subjects. The results are noteworthy, suggesting that NYCTALE outperforms the benchmark accuracy even with approximately 60% less training data on this demanding and small dataset.

NYCTALE: Neuro-Evidence Transformer for Adaptive and Personalized Lung Nodule Invasiveness Prediction

TL;DR

NYCTALE introduces a neuro-inspired evidence-accumulation Transformer for adaptive, slice-by-slice prediction of NSCLC invasiveness from CT sequences. The framework comprises a Shifting Window Transformer-based Evidence Encoder and a Drift-Diffusion Model–like Evidence Accumulation Module, outputting a decision only after accumulating sufficient evidence. On an in-house dataset of 114 SSNs, NYCTALE demonstrates competitive accuracy (~81–82%) and AUC (~0.77) while using substantially fewer slices than conventional pooling-based methods, highlighting data efficiency and potential interpretability advantages. By integrating cognitive neuroscience principles with transformer architectures, the approach advances personalized, efficient, and potentially more explainable lung cancer diagnosis within the PM paradigm.

Abstract

Drawing inspiration from the primate brain's intriguing evidence accumulation process, and guided by models from cognitive psychology and neuroscience, the paper introduces the NYCTALE framework, a neuro-inspired and evidence accumulation-based Transformer architecture. The proposed neuro-inspired NYCTALE offers a novel pathway in the domain of Personalized Medicine (PM) for lung cancer diagnosis. In nature, Nyctales are small owls known for their nocturnal behavior, hunting primarily during the darkness of night. The NYCTALE operates in a similarly vigilant manner, i.e., processing data in an evidence-based fashion and making predictions dynamically/adaptively. Distinct from conventional Computed Tomography (CT)-based Deep Learning (DL) models, the NYCTALE performs predictions only when sufficient amount of evidence is accumulated. In other words, instead of processing all or a pre-defined subset of CT slices, for each person, slices are provided one at a time. The NYCTALE framework then computes an evidence vector associated with contribution of each new CT image. A decision is made once the total accumulated evidence surpasses a specific threshold. Preliminary experimental analyses conducted using a challenging in-house dataset comprising 114 subjects. The results are noteworthy, suggesting that NYCTALE outperforms the benchmark accuracy even with approximately 60% less training data on this demanding and small dataset.
Paper Structure (9 sections, 2 equations, 5 figures, 1 table)

This paper contains 9 sections, 2 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Works
  3. The $\text{NYCTALE}$ Framework
  4. Data Description
  5. Data Preparation
  6. Evidence Encoder
  7. Evidence Accumulation Module
  8. Experimental Results
  9. Conclusion

Figures (5)

  • Figure 1: Sequence of sample slices in a CT volume. The yellow dashed rectangle highlights the middle slice.
  • Figure 2: Architecture of the $\text{NYCTALE}$ framework.
  • Figure 3: Model performance across different threshold values with slice addition direction of left to right.
  • Figure 4: (a) Amount of training data (total number of selected slices in fold $1$) needed in the last epoch for different threshold values. (b) Model performance (fold $1$) across different threshold values with slice addition direction of middle to the sides.
  • Figure 5: Histogram of training slices in fold $1$ across various subjects. (a) Distribution of slice numbers in the original training dataset, (b) Distribution of selected slices used for the last epoch of the training phase .