Evaluation of conventional and deep learning based image harmonization methods in radiomics studies

Objective: To evaluate the impact of image harmonization on outcome prediction models using radiomics. Approach: 234 patients from the Brain Tumor Image Segmentation Benchmark (BRATS) dataset with T1 MRI were enrolled in this study. Images were harmonized through a reference image using histogram matching (HHM) and a generative adversarial network (GAN)-based method (HGAN). 88 radiomics features were extracted on HHM, HGAN and original (HNONE) images. Wilcoxon paired test was used to identify features significantly impacted by the harmonization protocol used. Radiomic prediction models were built using feature selection with the Least Absolute Shrinkage and Selection Operator (LASSO) and Kaplan-Meier analysis. Main results: More than 50% of the features (49/88) were statistically modified by the harmonization with HHM and 55 with HGAN (adjusted p-value < 0.05). The contribution of histogram and texture features selected by the LASSO, in comparison to shape features that were not impacted by harmonization, was higher in harmonized datasets (47% for Hnone, 62% for HHM and 71% for HGAN). Both image-based harmonization methods allowed to split patients into two groups with significantly different survival (p<0.05). With the HGAN images, we were also able to build and validate a model using only features impacted by the harmonization (median survivals of 189 vs. 437 days, p=0.006) Significance: Data harmonization in a multi-institutional cohort allows to recover the predictive value of some radiomics features that was lost due to differences in the image properties across centers. In terms of ability to build survival prediction models in the BRATS dataset, the loss of power from impacted histogram and heterogeneity features was compensated by the selection of additional shape features. The harmonization using a GAN-based approach outperformed the histogram matching technique, supporting the interest for the development of new advanced harmonization techniques for radiomic analysis purposes.

Histogram-derived modified thresholds for coronary artery calcium scoring with lower tube voltage

We aimed to determine the proper modified thresholds for detecting and weighting CAC scores at 100 kV through histogram matching in comparison with 120 kV as a standard reference. From the training set (680 participants), modified thresholds at 100 kV were obtained through histogram matching of calcium pixels to 120 kV. From the validation set (213 participants), a standard CAC score at 120 kV, and modified CAC score at 100 kV using modified thresholds were compare through the paired t test and the Bland–Altman plot. Agreement for risk categories (no, minimal, mild, moderate, and severe) was evaluated using kappa statistics. Radiation doses were also compared. For the validation set, there was no significant difference between standard (median, 18.7; IQR, 0.0–207.0) and modified (median, 17.3; IQR, 0.0–220.9) CAC scores (P = 0.689). A small bias was achieved (0.74) with 95% limits of agreement from − 52.35 to 53.83. Agreements for risk categories were excellent (κ = 0.994). The mean dose-length-product of 100-kV scanning (30.1 ± 0.8 mGy * cm) was significantly decreased compared to 120-kV scanning (42.9 ± 0.6 mGy * cm) (P < 0.001). Histogram-derived modified thresholds at 100 kV can enable accurate CAC scoring while reducing radiation exposure.

Target Detection and Classification Improvements using Contrast Enhanced 16-bit Infrared Videos

In our earlier target detection and classification papers, we used 8-bit infrared videos in the Defense Systems Information Analysis Center(DSIAC) video dataset. In this paper, we focus on how we can improve the target detection and classification results using 16-bit videos. One problem with the 16-bit videos is that some image frames have very low contrast. Two methods were explored to improve upon previous detection and classification results. The first method used to improve contrast was effectively the same as the baseline 8-bit video data but using the 16-bit raw data rather than the 8-bit data taken from the avi files. The second method used was a second order histogram matching algorithm that preserves the 16-bit nature of the videos while providing normalization and contrast enhancement. Results showed the second order histogram matching algorithm improved the target detection using You Only Look Once (YOLO) and classificationusing Residual Network (ResNet) performance. The average precision (AP) metric in YOLO was improved by 8%. This is quite significant. The overall accuracy (OA) of ResNet has been improved by 12%. This is also very significant.

Evaluating the Effect of Intensity Standardisation on Longitudinal Whole Brain Atrophy Quantification in Brain Magnetic Resonance Imaging

Atrophy quantification is fundamental for understanding brain development and diagnosing and monitoring brain diseases. FSL-SIENA is a well-known fully automated method that has been widely used in brain magnetic resonance imaging studies. However, intensity variations arising during image acquisition may compromise evaluation, analysis and even diagnosis. In this work, we studied whether intensity standardisation could improve longitudinal atrophy quantification using FSL-SIENA. We evaluated the effect of six intensity standardisation methods—z-score, fuzzy c-means, Gaussian mixture model, kernel density estimation, histogram matching and WhiteStripe—on atrophy detected by FSL-SIENA. First, we evaluated scan–rescan repeatability using scans taken during the same session from OASIS (n=122). Except for WhiteStripe, intensity standardisation did not compromise the scan–rescan repeatability of FSL-SIENA. Second, we compared the mean annual atrophy for Alzheimer’s and control subjects from OASIS (n=122) and ADNI (n=147) yielded by FSL-SIENA with and without intensity standardisation, after adjusting for covariates. Our findings were threefold: First, the use of histogram matching was counterproductive, primarily as its assumption of equal tissue proportions does not necessarily hold in longitudinal studies. Second, standardising with z-score and WhiteStripe before registration affected the registration performance, thus leading to erroneous estimates. Third, z-score was the only method that consistently led to increased effect sizes compared to when omitted (no standardisation: 0.39 and 0.43 for OASIS and ADNI; z-score: 0.45 for both datasets). Overall, we found that incorporating z-score right after registration led to reduced inter-subject inter-scan intensity variability and benefited FSL-SIENA. Our work evinces the relevance of appropriate intensity standardisation in longitudinal cerebral atrophy assessments using FSL-SIENA.

Bleach correction ImageJ plugin for compensating the photobleaching of time-lapse sequences

During the capturing of the time-lapse sequence of fluorescently labeled samples, fluorescence intensity exhibits decays. This phenomenon is known as 'photobleaching' and is a widely known problem in imaging in life sciences. The photobleaching can be attenuated by tuning the imaging set-up, but when such adjustments only partially work, the image sequence can be corrected for the loss of intensity in order to precisely segment the target structure or to quantify true intensity dynamics. We implemented an ImageJ plugin that allows the user to compensate for the photobleaching to estimate the non-bleaching condition with choice of three different algorithms: simple ratio, exponential fitting, and histogram matching methods. The histogram matching method is a novel algorithm for photobleaching correction. This article presents details and characteristics of each algorithm based on application to actual image sequences.

MULTITEMPORAL HISTOGRAM MATCHING – A NEW APPROACH OF MOSS AND LICHEN CHANGE DETECTION FROM LANDSAT IN DATA-POOR ANTARCTICA ENVIRONMENTS

Mosses and lichens are important components of Antarctic ecosystems. Maps of these vegetation are needed to improve our understanding of ecosystem dynamics. This requires species distribution to be mapped repeatedly over time, a critical task that becomes extremely challenging in data-poor Antarctic regions, where the lack of field data, logistics, coupled with scarcity of cloud free, quality multitemporal Landsat imagery are major intrinsic constraints to time-series analysis for change detection. This study firstly analyzes the spectral curves of moss and lichen generated by field-based spectroradiometer and then proposes an innovative histogram matching technique where historical Landsat data is modified such that its histogram matches that of present (reference) dataset. This has made it possible to mapping multitemporal Landsat data in the Antarctic Peninsula. The results demonstrate an overall accuracy of 90.5%. Mapping of Arctic vegetation facilitated by histogram matching of Landsat image, according to the results, seems to be an advisable image processing technique for application in a data-poor context.