Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Preoperative Prognosis Assessment of Lumbar Spinal Surgery for Low Back Pain and Sciatica Patients based on Multimodalities and Multimodal Learning

Li-Chin Chen, Jung-Nien Lai, Hung-En Lin, Hsien-Te Chen, Kuo-Hsuan Hung, Yu Tsao

TL;DR

The study tackles the challenge of predicting preoperative outcomes for lumbar spinal surgery in LBP and sciatica by integrating multimodal data—standard clinical assessments, Traditional Chinese Medicine body constitution, planned surgical approach, and vowel pronunciation—within a multimodal learning framework. It systematically compares uni- and multi-modal configurations across early, joint, and late fusion strategies, using tabular data, free text, and acoustic features processed by CNNs, PubMedBERT-based NLP, and LSTM encoders, with CatBoost and other tree methods for unimodal tabular data. The key finding is that a combination of tabular data, text, and surgical plan information achieves the best accuracy (0.81) and AUROC (0.83) in a 105-patient cohort, with joint fusion offering robust performance and interpretable signals via IG and CCA analyses. The work demonstrates a practical, light-weight preoperative prognosis tool and provides actionable insights into modality contributions and fusion strategies for clinical decision support, bridging Eastern medicine concepts with modern ML techniques. Future work should scale sample size, refine audio features, and further validate interpretability in clinical settings.

Abstract

Low back pain (LBP) and sciatica may require surgical therapy when they are symptomatic of severe pain. However, there is no effective measures to evaluate the surgical outcomes in advance. This work combined elements of Eastern medicine and machine learning, and developed a preoperative assessment tool to predict the prognosis of lumbar spinal surgery in LBP and sciatica patients. Standard operative assessments, traditional Chinese medicine body constitution assessments, planned surgical approach, and vowel pronunciation recordings were collected and stored in different modalities. Our work provides insights into leveraging modality combinations, multimodals, and fusion strategies. The interpretability of models and correlations between modalities were also inspected. Based on the recruited 105 patients, we found that combining standard operative assessments, body constitution assessments, and planned surgical approach achieved the best performance in 0.81 accuracy. Our approach is effective and can be widely applied in general practice due to simplicity and effective.

Preoperative Prognosis Assessment of Lumbar Spinal Surgery for Low Back Pain and Sciatica Patients based on Multimodalities and Multimodal Learning

TL;DR

The study tackles the challenge of predicting preoperative outcomes for lumbar spinal surgery in LBP and sciatica by integrating multimodal data—standard clinical assessments, Traditional Chinese Medicine body constitution, planned surgical approach, and vowel pronunciation—within a multimodal learning framework. It systematically compares uni- and multi-modal configurations across early, joint, and late fusion strategies, using tabular data, free text, and acoustic features processed by CNNs, PubMedBERT-based NLP, and LSTM encoders, with CatBoost and other tree methods for unimodal tabular data. The key finding is that a combination of tabular data, text, and surgical plan information achieves the best accuracy (0.81) and AUROC (0.83) in a 105-patient cohort, with joint fusion offering robust performance and interpretable signals via IG and CCA analyses. The work demonstrates a practical, light-weight preoperative prognosis tool and provides actionable insights into modality contributions and fusion strategies for clinical decision support, bridging Eastern medicine concepts with modern ML techniques. Future work should scale sample size, refine audio features, and further validate interpretability in clinical settings.

Abstract

Low back pain (LBP) and sciatica may require surgical therapy when they are symptomatic of severe pain. However, there is no effective measures to evaluate the surgical outcomes in advance. This work combined elements of Eastern medicine and machine learning, and developed a preoperative assessment tool to predict the prognosis of lumbar spinal surgery in LBP and sciatica patients. Standard operative assessments, traditional Chinese medicine body constitution assessments, planned surgical approach, and vowel pronunciation recordings were collected and stored in different modalities. Our work provides insights into leveraging modality combinations, multimodals, and fusion strategies. The interpretability of models and correlations between modalities were also inspected. Based on the recruited 105 patients, we found that combining standard operative assessments, body constitution assessments, and planned surgical approach achieved the best performance in 0.81 accuracy. Our approach is effective and can be widely applied in general practice due to simplicity and effective.
Paper Structure (21 sections, 6 equations, 9 figures, 9 tables)

This paper contains 21 sections, 6 equations, 9 figures, 9 tables.

Table of Contents

  1. INTRODUCTION
  2. Multimodal fusion in biomedical fields
  3. MATERIALS AND METHODS
  4. Patient recruitment
  5. Standard operative assessments and planned surgical approach
  6. Traditional Chinese medicine assessment
  7. Vowel pronunciation
  8. Prognosis determination
  9. Uni- and multimodal fusion design
  10. Tabular data
  11. Free text data
  12. Audio vowel recordings
  13. Multimodal fusion design
  14. Training process and performance evaluation
  15. Interrelationship and interpretability
  16. ...and 6 more sections

Figures (9)

  • Figure 1: Patient assessments and ground truth determinations. BMI: Body mass index; VAS: Visual Analog Scale; EQ-5D: EuroQol Five Dimensions; ODI: Oswestry Disability Index.
  • Figure 2: Architecture design of uni-model for each modality. Conv1D: one-dimensional convolutional neural network; LeakyReLU: leaky rectified linear unit; FC: fully connected; LSTM: long short-term memory; $Pre$: preprocessor; $En$: encoder; $C$: classifier; $log1p$: natural logarithm of one plus the input.
  • Figure 3: Architecture of multimodal in early fusion design. $Pre$: preprocessor; $CNN$: convolution layer; Conv1D: one-dimensional convolutional neural network; LeakyReLU: leaky rectified linear unit; $En$: encoder; FC: fully connected; ReLU: rectified linear unit; $C$: classifier; $\bigoplus$: concatenation operation.
  • Figure 4: Architecture of proposed multimodal in joint fusion design. $Pre$: preprocessor; $En$: encoder; $C$: classifier; $\bigoplus$: concatenation operation.
  • Figure 5: Late fusion design for triple modalities. $En$: encoder, refers to any kind of algorithm; $\bigoplus$: concatenation operation
  • ...and 4 more figures