基于高分辨率 3D 微 CT 微钙化图像的分辨率对乳腺癌诊断的影响（模拟）。

The impact of (simulated) resolution on breast cancer diagnosis based on high-resolution 3D micro-CT microcalcification images.

机构信息

Department of Electronics and Informatics (ETRO), Vrije Universiteit Brussel (VUB), Brussels, Belgium.

Department of Radiology, Universitair Ziekenhuis Leuven, KU Leuven, Leuven, Belgium.

出版信息

Med Phys. 2024 Mar;51(3):1754-1762. doi: 10.1002/mp.16708. Epub 2023 Sep 12.

DOI:10.1002/mp.16708

PMID:37698346

Abstract

BACKGROUND

Breast microcalcifications (MCs) are considered to be a robust marker of breast cancer. A machine learning model can provide breast cancer diagnosis based on properties of individual MCs - if their characteristics are captured at high resolution and in 3D.

PURPOSE

The main purpose of the study was to explore the impact of image resolution (8 µm, 16 µm, 32 µm, 64 µm) when diagnosing breast cancer using radiomics features extracted from individual high resolution 3D micro-CT MC images.

METHODS

Breast MCs extracted from 86 female patients were analyzed at four different spatial resolutions: 8 µm (original resolution) and 16 µm, 32 µm, 64 µm (simulated image resolutions). Radiomic features were extracted at each image resolution in an attempt, to find a compact feature signature allowing to distinguish benign and malignant MCs. Machine learning algorithms were used for classifying individual MCs and samples (i.e., patients). For sample diagnosis, a custom-based thresholding approach was used to combine individual MC results into sample results. We conducted classification experiments when using (a) the same MCs visible in 8 µm, 16 µm, 32 µm, and 64 µm resolution; (b) the same MCs visible in 8 µm, 16 µm, and 32 µm resolution; (c) the same MCs visible in 8 µm and 16 µm resolution; (d) all MCs visible in 8 µm, 16 µm, 32 µm, and 64 µm resolution. Accuracy, sensitivity, specificity, AUC, and F1 score were computed for each experiment.

RESULTS

The individual MC results yielded an accuracy of 77.27%, AUC of 83.83%, F1 score of 77.25%, sensitivity of 80.86%, and specificity of 72.2% at 8 µm resolution. For the individual MC classifications we report for the F1 scores: a 2.29% drop when using 16 µm instead of 8 µm, a 4.01% drop when using 32 µm instead of 8 µm, a 10.69% drop when using 64 µm instead of 8 µm. The sample results yielded an accuracy and F1 score of 81.4%, sensitivity of 80.43%, and specificity value of 82.5% at 8 µm. For the sample classifications we report for F1 score values: a 6.3% drop when using 16 µm instead of 8 µm, a 4.91% drop when using 32 µm instead of 8 µm, and a 6.3% drop when using 64 µm instead of 8 µm.

CONCLUSIONS

The highest classification results are obtained at the highest resolution (8 µm). If breast MCs characteristics could be visualized/captured in 3D at a higher resolution compared to what is used nowadays in digital mammograms (approximately 70 µm), breast cancer diagnosis will be improved.

摘要

背景

乳腺微钙化（MCs）被认为是乳腺癌的一个强有力的标志物。机器学习模型可以根据个体 MC 的特性提供乳腺癌诊断-如果它们的特征以高分辨率和 3D 形式捕获。

目的

本研究的主要目的是探讨在使用从高分辨率 3D 微 CT MC 图像中提取的放射组学特征诊断乳腺癌时，图像分辨率（8 µm、16 µm、32 µm、64 µm）的影响。

方法

对 86 名女性患者的乳腺 MC 进行了分析，分辨率为 8 µm（原始分辨率）和 16 µm、32 µm、64 µm（模拟图像分辨率）。在每个图像分辨率中提取放射组学特征，试图找到一个紧凑的特征签名，允许区分良性和恶性 MC。使用机器学习算法对个体 MC 和样本（即患者）进行分类。对于样本诊断，使用基于定制的阈值方法将个体 MC 结果组合为样本结果。当使用（a）在 8 µm、16 µm、32 µm 和 64 µm 分辨率下可见的相同 MC；（b）在 8 µm、16 µm 和 32 µm 分辨率下可见的相同 MC；（c）在 8 µm 和 16 µm 分辨率下可见的相同 MC；（d）在 8 µm、16 µm、32 µm 和 64 µm 分辨率下可见的所有 MC 时，我们进行了分类实验。为每个实验计算了准确性、灵敏度、特异性、AUC 和 F1 分数。

结果

在 8 µm 分辨率下，个体 MC 结果的准确性为 77.27%，AUC 为 83.83%，F1 分数为 77.25%，灵敏度为 80.86%，特异性为 72.2%。对于我们报告的个体 MC 分类的 F1 分数：使用 16 µm 而不是 8 µm 时下降 2.29%，使用 32 µm 而不是 8 µm 时下降 4.01%，使用 64 µm 而不是 8 µm 时下降 10.69%。在 8 µm 时，样本结果的准确性和 F1 分数为 81.4%，灵敏度为 80.43%，特异性为 82.5%。对于我们报告的样本分类的 F1 分数值：使用 16 µm 而不是 8 µm 时下降 6.3%，使用 32 µm 而不是 8 µm 时下降 4.91%，使用 64 µm 而不是 8 µm 时下降 6.3%。