Physical and Engineering Sciences in Medicine最新文献

Identifying unknown types of diseases is a crucial step in preceding retinal imaging classification for the sake of safety, which is known as anomaly detection of retinal imaging. However, the widely-used supervised learning algorithms are not suitable for this problem, since the data of the unknown category is unobtainable. Moreover, for retinal imaging with different types of anomalous regions, using a single-resolution input causes information loss. Therefore, we propose an unsupervised auto-encoder model with multi-resolution inputs and outputs. We provide a theoretical understanding of the effectiveness of reconstruction error and the improvement of self-supervised learning for anomaly detection. Our experiments on two widely-used retinal imaging datasets show that the proposed methods are superior to other methods, and further experiments verify the validity of each part of the proposed method.

Sufficient dose reduction may not be achieved if radioprotective curtains are folded. This study aimed to evaluate the scattered dose rate distribution and physician eye lens dose at different curtain lengths. Using an over-couch fluoroscopy system, dH*(10)/dt was measured using a survey meter 150 cm from the floor at 29 positions in the examination room when the curtain lengths were 0% (no curtain), 50%, 75%, and 100%. The absorbed dose rates in the air at the positions of endoscopist and assistant were calculated using a Monte Carlo simulation by varying the curtain length from 0 to 100%. The air kerma was measured by 10 min fluoroscopy using optically stimulated luminescence dosimeters at the eye surfaces of the endoscopist phantom and the outside and inside of the radioprotective goggles. At curtain lengths of 50%, 75%, and 100%, the ratios of dH*(10)/dt relative to 0% ranged from 80.8 to 104.1%, 10.5 to 61.0%, and 11.8 to 24.8%, respectively. In the simulation, the absorbed dose rates at the endoscopist's and assistant's positions changed rapidly between 55 and 75% and 65% and 80% of the curtain length, respectively. At the 0%, 50%, 75%, and 100% curtain lengths, the air kerma at the left eye surface of the endoscopist phantom was 237 ± 29, 271 ± 30, 37.7 ± 7.5, and 33.5 ± 6.1 μGy, respectively. Therefore, a curtain length of 75% or greater is required to achieve a sufficient eye lens dose reduction effect at the position of the endoscopist.

Intensity-modulated radiation therapy (IMRT) has become a popular choice for breast cancer treatment. We aimed to evaluate and compare the robustness of each optimization method used for breast IMRT using TomoTherapy. A retrospective analysis was performed on 10 patients with left breast cancer. For each optimization method (clipping, virtual bolus, and skin flash), a corresponding 50 Gy/25 fr plan was created in the helical and direct TomoTherapy modes. The dose-volume histogram parameters were compared after shifting the patients anteriorly and posteriorly. In the helical mode, when the patient was not shifted, the median D1cc (minimum dose delivered to 1 cc of the organ volume) of the breast skin for the clipping and virtual bolus plans was 52.2 (interquartile range: 51.9-52.6) and 50.4 (50.1-50.8) Gy, respectively. After an anterior shift, D1cc of the breast skin for the clipping and virtual bolus plans was 56.0 (55.6-56.8) and 50.9 (50.5-51.3) Gy, respectively. When the direct mode was used without shifting the patient, D1cc of the breast skin for the clipping, virtual bolus, and skin flash plans was 52.6 (51.9-53.1), 53.4 (52.6-53.9), and 52.3 (51.7-53.0) Gy, respectively. After shifting anteriorly, D1cc of the breast skin for the clipping, virtual bolus, and skin flash plans was 55.6 (54.1-56.4), 52.4 (52.0-53.0), and 53.6 (52.6-54.6) Gy, respectively. The clipping method is not sufficient for breast IMRT. The virtual bolus and skin flash methods were more robust optimization methods according to our analyses.

Early diagnosis of prostate cancer, the most common malignancy in men, can improve patient outcomes. Since the tissue sampling procedures are invasive and sometimes inconclusive, an alternative image-based method can prevent possible complications and facilitate treatment management. We aim to propose a machine-learning model for tumor grade estimation based on ⁶⁸ Ga-PSMA-11 PET/CT images in prostate cancer patients. This study included 90 eligible participants out of 244 biopsy-proven prostate cancer patients who underwent staging ⁶⁸Ga-PSMA-11 PET/CT imaging. The patients were divided into high and low-intermediate groups based on their Gleason scores. The PET-only images were manually segmented, both lesion-based and whole prostate, by two experienced nuclear medicine physicians. Four feature selection algorithms and five classifiers were applied to Combat-harmonized and non-harmonized datasets. To evaluate the model's generalizability across different institutions, we performed leave-one-center-out cross-validation (LOOCV). The metrics derived from the receiver operating characteristic curve were used to assess model performance. In the whole prostate segmentation, combining the ANOVA algorithm as the feature selector with Random Forest (RF) and Extra Trees (ET) classifiers resulted in the highest performance among the models, with an AUC of 0.78 and 083, respectively. In the lesion-based segmentation, the highest AUC was achieved by MRMR feature selector + Linear Discriminant Analysis (LDA) and Logistic Regression (LR) classifiers (0.76 and 0.79, respectively). The LOOCV results revealed that both the RF_ANOVA and ET_ANOVA models showed high levels of accuracy and generalizability across different centers, with an average AUC value of 0.87 for the ET_ANOVA combination. Machine learning-based analysis of radiomics features extracted from ⁶⁸Ga-PSMA-11 PET/CT scans can accurately classify prostate tumors into low-risk and intermediate- to high-risk groups.

Ultrasound guidance has become the gold standard for obtaining vascular access. Angle information, which indicates the entry angle of the needle into the vein, is required to ensure puncture success. Although various image processing-based methods, such as deep learning, have recently been applied to improve needle visibility, these methods have limitations, in that the puncture angle to the target organ is not measured. We aim to detect the target vessel and puncture needle and to derive the puncture angle by combining deep learning and conventional image processing methods such as the Hough transform. Median cubital vein US images were obtained from 20 healthy volunteers, and images of simulated blood vessels and needles were obtained during the puncture of a simulated blood vessel in four phantoms. The U-Net architecture was used to segment images of blood vessels and needles, and various image processing methods were employed to automatically measure angles. The experimental results indicated that the mean dice coefficients of median cubital veins, simulated blood vessels, and needles were 0.826, 0.931, and 0.773, respectively. The quantitative results of angular measurement showed good agreement between the expert and automatic measurements of the puncture angle with 0.847 correlations. Our findings indicate that the proposed method achieves extremely high segmentation accuracy and automated angular measurements. The proposed method reduces the variability and time required in manual angle measurements and presents the possibility where the operator can concentrate on delicate techniques related to the direction of the needle.

In the realm of high-intensity focused ultrasound (HIFU) therapy, the precise prediction of lesion size during treatment planning remains a challenge, primarily due to the difficulty in quantitatively assessing energy deposition at the target site and the acoustic properties of the tissue through which the ultrasound wave propagates. This study investigates the hypothesis that the echo amplitude originating from the focus is indicative of acoustic attenuation and is directly related to the resultant lesion size. Echoes from multi-layered tissues, specifically porcine tenderloin and bovine livers, with varying fat thickness from 0 mm to 35 mm were collected using a focused ultrasound (FUS) transducer operated at a low power output and short duration. Subsequent to HIFU treatment under clinical conditions, the resulting lesion areas in the ex vivo tissues were meticulously quantified. A novel treatment strategy that prioritizes treatment spots based on descending echo amplitudes was proposed and compared with the conventional raster scan approach. Our findings reveal a consistent trend of decreasing echo amplitudes and HIFU-induced lesion areas with the increasing fat thickness. For porcine tenderloin, the values decreased from 2541.7 ± 641.9 mV and 94.4 ± 17.9 mm² to 385(342.5) mV and 24.9 ± 18.7 mm², and for bovine liver, from 1406(1202.5) mV and 94.4 ± 17.9 mm² to 502.1 ± 225.7 mV and 9.4 ± 6.3 mm², respectively, as the fat thickness increases from 0 mm to 35 mm. Significant correlations were identified between preoperative echo amplitudes and the HIFU-induced lesion areas (R = 0.833 and 0.784 for the porcine tenderloin and bovine liver, respectively). These correlations underscore the potential for an accurate and dependable prediction of treatment outcomes. Employing the proposed treatment strategy, the ex vivo experiment yielded larger lesion areas in bovine liver at a penetration depth of 8 cm compared to the conventional approach (58.84 ± 17.16 mm² vs. 44.28 ± 15.37 mm², p < 0.05). The preoperative echo amplitude from the FUS transducer is shown to be a reflective measure of acoustic attenuation within the wave propagation window and is closely correlated with the induced lesion areas. The proposed treatment strategy demonstrated enhanced efficiency in ex vivo settings, affirming the feasibility and accuracy of predicting HIFU-induced lesion size based on echo amplitude.

Hemodynamic parameters derived from pulse wave analysis have been shown to predict long-term outcomes in patients with heart failure (HF). Here we aimed to develop a deep-learning based algorithm that incorporates pressure waveforms for the identification and risk stratification of patients with HF. The first study, with a case-control study design to address data imbalance issue, included 431 subjects with HF exhibiting typical symptoms and 1545 control participants with no history of HF (non-HF). Carotid pressure waveforms were obtained from all the participants using applanation tonometry. The HF score, representing the probability of HF, was derived from a one-dimensional deep neural network (DNN) model trained with characteristics of the normalized carotid pressure waveform. In the second study of HF patients, we constructed a Cox regression model with 83 candidate clinical variables along with the HF score to predict the risk of all-cause mortality along with rehospitalization. To identify subjects using the HF score, the sensitivity, specificity, accuracy, F1 score, and area under receiver operating characteristic curve were 0.867, 0.851, 0.874, 0.878, and 0.93, respectively, from the hold-out cross-validation of the DNN, which was better than other machine learning models, including logistic regression, support vector machine, and random forest. With a median follow-up of 5.8 years, the multivariable Cox model using the HF score and other clinical variables outperformed the other HF risk prediction models with concordance index of 0.71, in which only the HF score and five clinical variables were independent significant predictors (p < 0.05), including age, history of percutaneous coronary intervention, concentration of sodium in the emergency room, N-terminal pro-brain natriuretic peptide, and hemoglobin. Our study demonstrated the diagnostic and prognostic utility of arterial waveforms in subjects with HF using a DNN model. Pulse wave contains valuable information that can benefit the clinical care of patients with HF.