# Improving Diagnostic Performance for Head and Neck Tumors with Simple Diffusion Kurtosis Imaging and Machine Learning Bi-Parameter Analysis

## TL;DR

This study shows that combining machine learning with diffusion imaging improves the accuracy of diagnosing head and neck tumors as benign or malignant.

## Contribution

The study introduces a machine learning-based bi-parameter analysis using SDI data for improved tumor diagnosis.

## Key findings

- Bi-parameter analysis with gradient boosting achieved an AUC of 0.81 in differentiating benign and malignant tumors.
- Using MK and ADC values from SDI with ML improved diagnostic performance compared to single-parameter methods.
- Pre-processing with a Gaussian filter enhanced the accuracy of tumor classification.

## Abstract

Background/Objectives: Mean kurtosis (MK) values in simple diffusion kurtosis imaging (SDI)—a type of diffusion kurtosis imaging (DKI)—have been reported to be useful in the diagnosis of head and neck malignancies, for which pre-processing with smoothing filters has been reported to improve the diagnostic accuracy. Multi-parameter analysis using DKI in combination with other image types has recently been reported to improve the diagnostic performance. The purpose of this study was to evaluate the usefulness of machine learning (ML)-based multi-parameter analysis using the MK and apparent diffusion coefficient (ADC) values—which can be acquired simultaneously through SDI—for the differential diagnosis of benign and malignant head and neck tumors, which is important for determining the treatment strategy, as well as examining the usefulness of filter pre-processing. Methods: A total of 32 pathologically diagnosed head and neck tumors were included in the study, and a Gaussian filter was used for image pre-processing. MK and ADC values were extracted from pixels within the tumor area and used as explanatory variables. Five ML algorithms were used to create models for the prediction of tumor status (benign or malignant), which were evaluated through ROC analysis. Results: Bi-parameter analysis with gradient boosting achieved the best diagnostic performance, with an AUC of 0.81. Conclusions: The usefulness of bi-parameter analysis with ML methods for the differential diagnosis of benign and malignant head and neck tumors using SDI data were demonstrated.