Enhancing Diagnostic Precision: Evaluation of Preprocessing Filters in Simple Diffusion Kurtosis Imaging for Head and Neck Tumors.

Background: Our initial clinical study using simple diffusion kurtosis imaging (SDI), which simultaneously produces a diffusion kurtosis image (DKI) and an apparent diffusion coefficient map, confirmed the usefulness of SDI for tumor diagnosis. However, the obtained DKI had noticeable variability in the mean kurtosis (MK) values, which is inherent to SDI. We aimed to improve this variability in SDI by preprocessing with three different filters (Gaussian [G], median [M], and nonlocal mean) of the diffusion-weighted images used for SDI. Methods: The usefulness of filter parameters for diagnosis was examined in basic and clinical studies involving 13 patients with head and neck tumors. Results: The filter parameters, which did not change the median MK value, but reduced the variability and significantly homogenized the MK values in tumor and normal tissues in both basic and clinical studies, were identified. In the receiver operating characteristic curve analysis for distinguishing tumors from normal tissues using MK values, the area under curve values significantly improved from 0.627 without filters to 0.641 with G (σ = 0.5) and 0.638 with M (radius = 0.5). Conclusions: Thus, image pretreatment with G and M for SDI was shown to be useful for improving tumor diagnosis in clinical practice.

Characteristic Mean Kurtosis Values in Simple Diffusion Kurtosis Imaging of Dentigerous Cysts.

We evaluated the usefulness of simple diffusion kurtosis (SD) imaging, which was developed to generate diffusion kurtosis images simultaneously with an apparent diffusion coefficient (ADC) map for 27 cystic disease lesions in the head and neck region. The mean kurtosis (MK) and ADC values were calculated for the cystic space. The MK values were dentigerous cyst (DC): 0.74, odontogenic keratocyst (OKC): 0.86, ranula (R): 0.13, and mucous cyst (M): 0, and the ADC values were DC: 1364 × 10-6 mm2/s, OKC: 925 × 10-6 mm2/s, R: 2718 × 10-6 mm2/s, and M: 2686 × 10-6 mm2/s. The MK values of DC and OKC were significantly higher than those of R and M, whereas their ADC values were significantly lower. One reason for the characteristic signal values in diffusion-weighted images of DC may be related to content components such as fibrous tissue and exudate cells. When imaging cystic disease in the head and neck region using SD imaging, the maximum b-value setting at the time of imaging should be limited to approximately 1200 s/mm2 for accurate MK value calculation. This study is the first to show that the MK values of DC are characteristically higher than those of other cysts.

Evaluation of the accuracy of heart dose prediction by machine learning for selecting patients not requiring deep inspiration breath­hold radiotherapy after breast cancer surgery.

Increased heart dose during postoperative radiotherapy (RT) for left-sided breast cancer (BC) can cause cardiac injury, which can decrease patient survival. The deep inspiration breath-hold technique (DIBH) is becoming increasingly common for reducing the mean heart dose (MHD) in patients with left-sided BC. However, treatment planning and DIBH for RT are laborious, time-consuming and costly for patients and RT staff. In addition, the proportion of patients with left BC with low MHD is considerably higher among Asian women, mainly due to their smaller breast volume compared with that in Western countries. The present study aimed to determine the optimal machine learning (ML) model for predicting the MHD after RT to pre-select patients with low MHD who will not require DIBH prior to RT planning. In total, 562 patients with BC who received postoperative RT were randomly divided into the trainval (n=449) and external (n=113) test datasets for ML using Python (version 3.8). Imbalanced data were corrected using synthetic minority oversampling with Gaussian noise. Specifically, right-left, tumor site, chest wall thickness, irradiation method, body mass index and separation were the six explanatory variables used for ML, with four supervised ML algorithms used. Using the optimal value of hyperparameter tuning with root mean squared error (RMSE) as an indicator for the internal test data, the model yielding the best F2 score evaluation was selected for final validation using the external test data. The predictive ability of MHD for true MHD after RT was the highest among all algorithms for the deep neural network, with a RMSE of 77.4, F2 score of 0.80 and area under the curve-receiver operating characteristic of 0.88, for a cut-off value of 300 cGy. The present study suggested that ML can be used to pre-select female Asian patients with low MHD who do not require DIBH for the postoperative RT of BC.

Mean Heart Dose Prediction Using Parameters of Single-Slice Computed Tomography and Body Mass Index: Machine Learning Approach for Radiotherapy of Left-Sided Breast Cancer of Asian Patients.

Deep inspiration breath-hold (DIBH) is an excellent technique to reduce the incidental radiation received by the heart during radiotherapy in patients with breast cancer. However, DIBH is costly and time-consuming for patients and radiotherapy staff. In Asian countries, the use of DIBH is restricted due to the limited number of patients with a high mean heart dose (MHD) and the shortage of radiotherapy personnel and equipment compared to that in the USA. This study aimed to develop, evaluate, and compare the performance of ten machine learning algorithms for predicting MHD using a patient's body mass index and single-slice CT parameters to identify patients who may not require DIBH. Machine learning models were built and tested using a dataset containing 207 patients with left-sided breast cancer who were treated with field-in-field radiotherapy with free breathing. The average MHD was 251 cGy. Stratified repeated four-fold cross-validation was used to build models using 165 training data. The models were compared internally using their average performance metrics: F2 score, AUC, recall, accuracy, Cohen's kappa, and Matthews correlation coefficient. The final performance evaluation for each model was further externally analyzed using 42 unseen test data. The performance of each model was evaluated as a binary classifier by setting the cut-off value of MHD ≥ 300 cGy. The deep neural network (DNN) achieved the highest F2 score (78.9%). Most models successfully classified all patients with high MHD as true positive. This study indicates that the ten models, especially the DNN, might have the potential to identify patients who may not require DIBH.

Usefulness of Simple Diffusion Kurtosis Imaging for Head and Neck Tumors: An Early Clinical Study.

Diffusion kurtosis (DK) imaging (DKI), a type of restricted diffusion-weighted imaging, has been reported to be useful for tumor diagnoses in clinical studies. We developed a software program to simultaneously create DK images with apparent diffusion coefficient (ADC) maps and conducted an initial clinical study. Multi-shot echo-planar diffusion-weighted images were obtained at b-values of 0, 400, and 800 sec/mm2 for simple DKI, and DK images were created simultaneously with the ADC map. The usefulness of the DK image and ADC map was evaluated using a pixel analysis of all pixels and a median analysis of the pixels of each case. Tumor and normal tissues differed significantly in both pixel and median analyses. In the pixel analysis, the area under the curve was 0.64 for the mean kurtosis (MK) value and 0.77 for the ADC value. In the median analysis, the MK value was 0.74, and the ADC value was 0.75. The MK and ADC values correlated moderately in the pixel analysis and strongly in the median analysis. Our simple DKI system created DK images simultaneously with ADC maps, and the obtained MK and ADC values were useful for differentiating head and neck tumors from normal tissue.

Quantitative evaluation of the reduction of distortion and metallic artifacts in magnetic resonance images using the multiacquisition variable­resonance image combination selective sequence.

Magnetic resonance imaging (MRI) is superior to computed tomography (CT) in determining changes in tissue structure, such as those observed following inflammation and infection. However, when metal implants or other metal objects are present, MRI exhibits more distortion and artifacts compared with CT, which hinders the accurate measurement of the implants. A limited number of reports have examined whether the novel MRI sequence, multiacquisition variable-resonance image combination selective (MAVRIC SL), can accurately measure metal implants without distortion. Therefore, the present study aimed to demonstrate whether MAVRIC SL could accurately measure metal implants without distortion and whether the area around the metal implants could be well delineated without artifacts. An agar phantom containing a titanium alloy lumbar implant was used for the present study and was imaged using a 3.0 T MRI machine. A total of three imaging sequences, namely MAVRIC SL, CUBE and magnetic image compilation (MAGiC), were applied and the results were compared. Distortion was evaluated by measuring the screw diameter and distance between the screws multiple times in the phase and frequency directions by two different investigators. The artifact region around the implant was examined using a quantitative method following standardization of the phantom signal values. It was revealed that MAVRIC SL was a superior sequence compared with CUBE and MAGiC, as there was significantly less distortion, a lack of bias between the two different investigators and significantly reduced artifact regions. These results suggested the possibility of utilizing MAVRIC SL for follow-up to observe metal implant insertions.

Evaluation of calculation processes of apparent diffusion coefficient subtraction method (ASM) imaging.

A number of restricted diffusion (RD) imaging techniques, such as diffusion kurtosis (DK) imaging and Q space imaging, have been developed and proven to be useful for the diagnosis of diseases, including cerebral gliomas and cerebrovascular infarction. In particular, apparent diffusion coefficient (ADC) subtraction method (ASM) imaging has become available recently as a novel RD imaging technique. ASM is based on the difference between the ADC values in an image pair of two ADC maps, ADC basic (ADCb) and ADC modify (ADCm), which are created from diffusion-weighted images taken using short and long effective diffusion times, respectively. The present study aimed to assess the potential of different types of ASM imaging by comparing them with DK imaging which is the gold-standard RD imaging technique. In the present basic study using both polyethylene glycol phantom and cell-containing bio-phantom, three different types of ASM images were created using different calculation processes. ASM/A is an image calculated by dividing the absolute difference between ADCb and ADCm by ADCb several times. By contrast, ASM/S is an image created by dividing the absolute difference between ADCb and ADCm by the standard deviation of ADCb several times. As for positive ASM/A image (PASM/A), the positive image, which was resultant after subtracting ADCb from ADCm, was divided by ADCb several times. A comparison was made between the types of ASM and DK images. The results showed the same tendency between ASM/A in addition to both ASM/S and PASM/A. By increasing the number of divisions by ADCb from three to five times, ASM/A images transformed from DK-mimicking to more RD-sensitive images compared with DK images. These observations suggest that ASM/A images may prove useful for future clinical applications in RD imaging protocols for the diagnosis of diseases.

Evaluation of Fast Diffusion Kurtosis Imaging Using New Software Designed for Widespread Clinical Use.

Clinical research using restricted diffusion-weighted imaging, especially diffusion kurtosis (DK) imaging, has been progressing, with reports on its effectiveness in the diagnostic imaging of cerebral infarctions, neurodegenerative diseases, and tumors, among others. However, the application of DK imaging in daily clinical practice has not spread because of the long imaging time required and the use of specific software for image creation. Herein, with the aim of promoting clinical research using DK imaging at any medical facility, we evaluated fast DK imaging using a new software program. We developed a new macro program that produces DK images using general-purpose, inexpensive software (Microsoft Excel and ImageJ), and we evaluated fast DK imaging using bio-phantoms and a healthy volunteer in clinical trials. The DK images created by the new software with diffusion-weighted images captured with short-time imaging sequences were similar to the original DK images captured with long-time imaging sequences. The DK images using three b-values, which can reduce the imaging time by 43%, were equivalent to the DK images using five b-values. The DK imaging technique developed herein might allow any medical facility to increase its daily clinical use of DK imaging and easily conduct clinical research.

Development and Evaluation of a Short-time Imaging Method for the Clinical Study of the Apparent Diffusion Coefficient Subtraction Method.

The apparent diffusion coefficient subtraction method (ASM) was developed as a new restricted diffusionweighted imaging technique for magnetic resonance imaging (MRI). The usefulness of the ASM has been established by in vitro basic research using a bio-phantom, and clinical research on the application of the ASM for the human body is needed. Herein, we developed a short-time sequence for ASM imaging of the heads of healthy volunteers (n=2), and we investigated the similarity between the obtained ASM images and diffusion kurtosis (DK) images to determine the utility of the ASM for clinical uses. This study appears to be the first to report ASM images of the human head. We observed that the short-time sequence for the ASM imaging of the head can be scanned in approx. 3 min at 1.5T MRI. The noise reduction effect of median filter processing was confirmed on the ASM images scanned by this sequence. The obtained ASM images showed a weak correlation with the DK images, indicating that the ASM images are restricted diffusion-weighted images. The new shorttime imaging sequence could thus be used in clinical studies applying the ASM.