Fast ICCD-based temperature modulated fluorescence tomography.

Fluorescence tomography (FT) has become a powerful preclinical imaging modality with a great potential for several clinical applications. Although it has superior sensitivity and utilizes low-cost instrumentation, the highly scattering nature of bio-tissue makes FT in thick samples challenging, resulting in poor resolution and low quantitative accuracy. To overcome the limitations of FT, we previously introduced a novel method, termed temperature modulated fluorescence tomography (TMFT), which is based on two key elements: (1) temperature-sensitive fluorescent agents (ThermoDots) and (2) high-intensity focused ultrasound (HIFU). The fluorescence emission of ThermoDots increases up to hundredfold with only several degree temperature elevation. The exceptional and reversible response of these ThermoDots enables their modulation, which effectively allows their localization using the HIFU. Their localization is then used as functional a priori during the FT image reconstruction process to resolve their distribution with higher spatial resolution. The last version of the TMFT system was based on a cooled CCD camera utilizing a step-and-shoot mode, which necessitated long total imaging time only for a small selected region of interest (ROI). In this paper, we present the latest version of our TMFT technology, which uses a much faster continuous HIFU scanning mode based on an intensified CCD (ICCD) camera. This new, to the best of our knowledge, version can capture the whole field-of-view (FOV) of 50×30m m 2 at once and reduces the total imaging time down to 30 min, while preserving the same high resolution (∼1.3m m) and superior quantitative accuracy (<7% error) as the previous versions. Therefore, this new method is an important step toward utilization of TMFT for preclinical imaging.

Prediction of breast cancer molecular subtypes on DCE-MRI using convolutional neural network with transfer learning between two centers.

OBJECTIVES: To apply deep learning algorithms using a conventional convolutional neural network (CNN) and a recurrent CNN to differentiate three breast cancer molecular subtypes on MRI. METHODS: A total of 244 patients were analyzed, 99 in training dataset scanned at 1.5 T and 83 in testing-1 and 62 in testing-2 scanned at 3 T. Patients were classified into 3 subtypes based on hormonal receptor (HR) and HER2 receptor: (HR+/HER2-), HER2+, and triple negative (TN). Only images acquired in the DCE sequence were used in the analysis. The smallest bounding box covering tumor ROI was used as the input for deep learning to develop the model in the training dataset, by using a conventional CNN and the convolutional long short-term memory (CLSTM). Then, transfer learning was applied to re-tune the model using testing-1(2) and evaluated in testing-2(1). RESULTS: In the training dataset, the mean accuracy evaluated using tenfold cross-validation was higher by using CLSTM (0.91) than by using CNN (0.79). When the developed model was applied to the independent testing datasets, the accuracy was 0.4-0.5. With transfer learning by re-tuning parameters in testing-1, the mean accuracy reached 0.91 by CNN and 0.83 by CLSTM, and improved accuracy in testing-2 from 0.47 to 0.78 by CNN and from 0.39 to 0.74 by CLSTM. Overall, transfer learning could improve the classification accuracy by greater than 30%. CONCLUSIONS: The recurrent network using CLSTM could track changes in signal intensity during DCE acquisition, and achieved a higher accuracy compared with conventional CNN during training. For datasets acquired using different settings, transfer learning can be applied to re-tune the model and improve accuracy. KEY POINTS: • Deep learning can be applied to differentiate breast cancer molecular subtypes. • The recurrent neural network using CLSTM could track the change of signal intensity in DCE images, and achieved a higher accuracy compared with conventional CNN during training. • For datasets acquired using different scanners with different imaging protocols, transfer learning provided an efficient method to re-tune the classification model and improve accuracy.

Development of U-Net Breast Density Segmentation Method for Fat-Sat MR Images Using Transfer Learning Based on Non-Fat-Sat Model.

To develop a U-net deep learning method for breast tissue segmentation on fat-sat T1-weighted (T1W) MRI using transfer learning (TL) from a model developed for non-fat-sat images. The training dataset (N = 126) was imaged on a 1.5 T MR scanner, and the independent testing dataset (N = 40) was imaged on a 3 T scanner, both using fat-sat T1W pulse sequence. Pre-contrast images acquired in the dynamic-contrast-enhanced (DCE) MRI sequence were used for analysis. All patients had unilateral cancer, and the segmentation was performed using the contralateral normal breast. The ground truth of breast and fibroglandular tissue (FGT) segmentation was generated using a template-based segmentation method with a clustering algorithm. The deep learning segmentation was performed using U-net models trained with and without TL, by using initial values of trainable parameters taken from the previous model for non-fat-sat images. The ground truth of each case was used to evaluate the segmentation performance of the U-net models by calculating the dice similarity coefficient (DSC) and the overall accuracy based on all pixels. Pearson's correlation was used to evaluate the correlation of breast volume and FGT volume between the U-net prediction output and the ground truth. In the training dataset, the evaluation was performed using tenfold cross-validation, and the mean DSC with and without TL was 0.97 vs. 0.95 for breast and 0.86 vs. 0.80 for FGT. When the final model developed with and without TL from the training dataset was applied to the testing dataset, the mean DSC was 0.89 vs. 0.83 for breast and 0.81 vs. 0.81 for FGT, respectively. Application of TL not only improved the DSC, but also decreased the required training case number. Lastly, there was a high correlation (R2 > 0.90) for both the training and testing datasets between the U-net prediction output and ground truth for breast volume and FGT volume. U-net can be applied to perform breast tissue segmentation on fat-sat images, and TL is an efficient strategy to develop a specific model for each different dataset.

Feasibility study of high spatial resolution multimodality fluorescence tomography in ex vivo biological tissue.

Previously, we demonstrated that temperature-modulated fluorescence tomography (TM-FT) could provide fluorescence images with high quantitative accuracy and the spatial resolution of focused ultrasound. TM-FT is based on scanning the focused ultrasound across the medium to activate temperature-reversible fluorescent nanoprobes (ThermoDots). This technique can resolve small fluorescent targets located several centimeters deep in turbid media with millimeter resolution. Our past studies with this multimodality technique used agar phantoms, which could not represent the true heterogeneous nature of the acoustic and optical properties of biological tissue. In this work, we report the results of the first TM-FT study performed on ex vivo chicken breast tissue. In order to improve the spatial resolution of this technique, diffuse optical tomography is also used to better estimate the optical property maps of the tissue, which is utilized as functional a priori for the TM-FT reconstruction algorithm. These ex vivo results show that TM-FT can accurately recover the concentration and position of a 1.5 mm×5 mm inclusion filled with ThermoDots. Since the inclusion is embedded 2 cm deep in the chicken breast sample, these results demonstrate the great potential of TM-FT for future in vivo small animal imaging.

Experimental evaluation of the resolution and quantitative accuracy of temperature-modulated fluorescence tomography.

Previously, we reported on the spatial resolution and quantitative accuracy of temperature-modulated fluorescence tomography (TM-FT) using simulation studies. TM-FT is a novel fully integrated multimodality imaging technique that combines fluorescence diffuse optical tomography (FT) with focused ultrasound. Utilizing unique thermo-reversible fluorescent nanocapsules (ThermoDots), TM-FT provides high-resolution cross-sectional fluorescence images in thick tissue (up to 6 cm). Focused ultrasound and temperature-sensitive ThermoDots are combined to provide accurate localization of these fluorescent probes and functional a priori information to constrain the conventional FT reconstruction algorithm. Our previous simulation studies evaluated the performance of TM-FT using synthetic phantoms with multiple fluorescence targets of various sizes located at different depths. In this follow-up work, we perform experimental studies to evaluate the performance of this hybrid imaging system, in particular, the effect of size, depth, and concentration of the fluorescence target. While FT alone is unable to accurately locate and resolve the fluorophore target in many cases, TM-FT is able to resolve the size and concentration of the ThermoDots within a thick turbid medium with high accuracy for all cases. The maximum error in the recovered ThermoDots concentration and target sizes with TM-FT are 12% and 25%, respectively.

Differentiation of tumor vasculature heterogeneity levels in small animals based on total hemoglobin concentration using magnetic resonance-guided diffuse optical tomography in vivo.

Insight into the vasculature of the tumor in small animals has the potential to impact many areas of cancer research. The heterogeneity of the vasculature of a tumor is directly related to tumor stage and disease progression. In this small scale animal study, we investigated the feasibility of differentiating tumors with different levels of vasculature heterogeneity in vivo using a previously developed hybrid magnetic resonance imaging (MRI) and diffuse optical tomography (DOT) system for small animal imaging. Cross-sectional total hemoglobin concentration maps of 10 Fisher rats bearing R3230 breast tumors are reconstructed using multi-wavelength DOT measurements both with and without magnetic resonance (MR) structural a priori information. Simultaneously acquired MR structural images are used to guide and constrain the DOT reconstruction, while dynamic contrast-enhanced MR functional images are used as the gold standard to classify the vasculature of the tumor into two types: high versus low heterogeneity. These preliminary results show that the stand-alone DOT is unable to differentiate tumors with low and high vascular heterogeneity without structural a priori information provided by a high resolution imaging modality. The mean total hemoglobin concentrations comparing the vasculature of the tumors with low and high heterogeneity are significant (p-value 0.02) only when MR structural a priori information is utilized.

Implementation of a new scanning method for high-resolution fluorescence tomography using thermo-sensitive fluorescent agents.

Conventional fluorescence tomography provides images of the distribution of fluorescent agents within highly scattering media, but suffers from poor spatial resolution. Previously, we introduced a new method termed "temperature-modulated fluorescence tomography" (TM-FT) that generates fluorescence images with high spatial resolution. TM-FT first uses focused ultrasound to locate the distribution of temperature-sensitive fluorescence probes. Afterward, this a priori information is utilized to improve the performance of the inverse solver for conventional fluorescence tomography and reveal quantitatively accurate fluorophore concentration maps. However, the disadvantage of this novel method is the long data acquisition time as the ultrasound beam was scanned in a step-and-shoot mode. In this Letter, we present a new, fast scanning method that reduces the imaging time 40 fold. By continuously scanning the ultrasound beam over a 50 mm by 25 mm field-of-view, high-resolution fluorescence images are obtained in less than 29 min, which is critical for in vivo small animal imaging.

Simulation-based evaluation of the resolution and quantitative accuracy of temperature-modulated fluorescence tomography.

Conventional fluorescence tomography (FT) can recover the distribution of fluorescent agents within a highly scattering medium. However, poor spatial resolution remains its foremost limitation. Previously, we introduced a new fluorescence imaging technique termed "temperature-modulated fluorescence tomography" (TM-FT), which provides high-resolution images of fluorophore distribution. TM-FT is a multimodality technique that combines fluorescence imaging with focused ultrasound to locate thermo-sensitive fluorescence probes using a priori spatial information to drastically improve the resolution of conventional FT. In this paper, we present an extensive simulation study to evaluate the performance of the TM-FT technique on complex phantoms with multiple fluorescent targets of various sizes located at different depths. In addition, the performance of the TM-FT is tested in the presence of background fluorescence. The results obtained using our new method are systematically compared with those obtained with the conventional FT. Overall, TM-FT provides higher resolution and superior quantitative accuracy, making it an ideal candidate for in vivo preclinical and clinical imaging. For example, a 4 mm diameter inclusion positioned in the middle of a synthetic slab geometry phantom (D:40 mm×W:100 mm) is recovered as an elongated object in the conventional FT (x=4.5 mm; y=10.4 mm), while TM-FT recovers it successfully in both directions (x=3.8 mm; y=4.6 mm). As a result, the quantitative accuracy of the TM-FT is superior because it recovers the concentration of the agent with a 22% error, which is in contrast with the 83% error of the conventional FT.

Development of a preclinical CCD-based temperature modulated fluorescence tomography platform.

In preclinical research, fluorescence molecular tomography (FMT) is the most sensitive imaging modality to interrogate whole-body and provide 3D distribution of fluorescent contract agents. Despite its superior sensitivity, its mediocre spatial-resolution has been the main barrier to its clinical translation. This limitation is mainly due to the high scattering of optical photons in biological tissue together with the limited boundary measurements that lead to an undetermined and ill-posed inverse problem. To overcome the limitations of FMT, we previously introduced a novel method termed, Temperature Modulated Fluorescence Tomography (TMFT). TMFT utilizes thermos-sensitive fluorescent agents (ThermoDots) as a key component and localizes them with high-intensity focused ultrasound (HIFU). Scanning the focused HIFU beam having a diameter Ø = 1.3 mm across the tissue while monitoring the variation in the measured fluorescence signals reveals the position of the ThermoDots with high spatial accuracy. We have formerly built a prototype TMFT system that uses optical fibers for detection. In this paper, we present an upgraded version using a CCD camera-based detection that enables non-contact imaging. In this version, the animal under investigation is placed on an ultrasound transparent membrane, which eliminates the need for its immersion in optical matching fluids that were required by the fiber-based system. This CCD-based system will pave the way for convenient and wide-spread use of TMFT in preclinical research. Its performance validation on phantom studies demonstrates that high spatial-resolution (∼1.3 mm) and quantitative accuracy in recovered fluorophore concentration (<3% error) can be achieved.

A thermo-sensitive fluorescent agent based method for excitation light leakage rejection for fluorescence molecular tomography.

Fluorescence molecular tomography (FMT) is widely used in preclinical oncology research. FMT is the only imaging technique able to provide 3D distribution of fluorescent probes within thick highly scattering media. However, its integration into clinical medicine has been hampered by its low spatial resolution caused by the undetermined and ill-posed nature of its reconstruction algorithm. Another major factor degrading the quality of FMT images is the large backscattered excitation light component leaking through the rejection filters and coinciding with the weak fluorescent signal arising from a low tissue fluorescence concentration. In this paper, we present a new method based on the use of a novel thermo-sensitive fluorescence probe. In fact, the excitation light leakage is accurately estimated from a set of measurements performed at different temperatures and then is corrected for in the tomographic data. The obtained results show a considerable improvement in both spatial resolution and quantitative accuracy of FMT images due to the proper correction of fluorescent signals.

Dosimetric comparisons of different hypofractionated stereotactic radiotherapy techniques in treating intracranial tumors > 3 cm in longest diameter.

OBJECTIVE: The authors sought to compare the dosimetric quality of hypofractionated stereotactic radiosurgery in treating sizeable brain tumors across the following treatment platforms: GammaKnife (GK) Icon, CyberKnife (CK) G4, volumetric modulated arc therapy (VMAT) on the Varian TrueBeam STx, double scattering proton therapy (DSPT) on the Mevion S250, and intensity modulated proton therapy (IMPT) on the Varian ProBeam. METHODS: In this retrospective study, stereotactic radiotherapy treatment plans were generated for 10 patients with sizeable brain tumors (> 3 cm in longest diameter) who had been treated with VMAT. Six treatment plans, 20-30 Gy in 5 fractions, were generated for each patient using the same constraints for each of the following radiosurgical methods: 1) GK, 2) CK, 3) coplanar arc VMAT (VMAT-C), 4) noncoplanar arc VMAT (VMAT-NC), 5) DSPT, and 6) IMPT. The coverage; conformity index; gradient index (GI); homogeneity index; mean and maximum point dose of organs at risk; total dose volume (V) in Gy to the normal brain for 2 Gy (V2), 5 Gy (V5), and 12 Gy (V12); and integral dose were compared across all platforms. RESULTS: Among the 6 techniques, GK consistently produced a sharper dose falloff despite a greater central target dose. GK gave the lowest GI, with a mean of 2.7 ± 0.1, followed by CK (2.9 ± 0.1), VMAT-NC (3.1 ± 0.3), and VMAT-C (3.5 ± 0.3). The highest mean GIs for the proton beam treatments were 3.8 ± 0.4 for DSPT and 3.9 ± 0.4 for IMPT. The GK consistently targeted the lowest normal brain volume, delivering 5 to 12 Gy when treating relatively smaller- to intermediate-sized lesions (less than 15-20 cm3). Yet, the differences across the 6 modalities relative to GK decreased with the increase of target volume. In particular, the proton treatments delivered the lowest V5 to the normal brain when the target size was over 15-20 cm3 and also produced the lowest integral dose to the normal brain regardless of the target size. CONCLUSIONS: This study provides an insightful understanding of dosimetric quality from both photon and proton treatment across the most advanced stereotactic radiotherapy platforms.

Machine learning for prediction of chemoradiation therapy response in rectal cancer using pre-treatment and mid-radiation multi-parametric MRI.

PURPOSE: To predict the neoadjuvant chemoradiation therapy (CRT) response in patients with locally advanced rectal cancer (LARC) using radiomics and deep learning based on pre-treatment MRI and a mid-radiation follow-up MRI taken 3-4 weeks after the start of CRT. METHODS: A total of 51 patients were included, 45 with pre-treatment, 41 with mid-radiation therapy (RT), and 35 with both MRI sets. The multi-parametric MRI protocol included T2, diffusion weighted imaging (DWI) with b-values of 0 and 800 s/mm2, and dynamic-contrast-enhanced (DCE) MRI. After completing CRT and surgery, the specimen was examined to determine the pathological response based on the tumor regression grade. The tumor ROI was manually drawn on the post-contrast image and mapped to other sequences. The total tumor volume and mean apparent diffusion coefficient (ADC) were measured. Radiomics using GLCM texture and histogram parameters, and deep learning using a convolutional neural network (CNN), were performed to differentiate pathologic complete response (pCR) vs. non-pCR, and good response (GR) vs. non-GR. RESULTS: Tumor volume decreased and ADC increased significantly in the mid-RT MRI compared to the pre-treatment MRI. For predicting pCR vs. non-pCR, combining ROI and radiomics features achieved an AUC of 0.80 for pre-treatment, 0.82 for mid-RT, and 0.86 for both MRI together. For predicting GR vs. non-GR, the AUC was 0.91 for pre-treatment, 0.92 for mid-RT, and 0.93 for both MRI together. In deep learning using CNN, combining pre-treatment and mid-RT MRI achieved a higher accuracy compared to using either dataset alone, with AUC of 0.83 for predicting pCR vs. non-pCR. CONCLUSION: Radiomics based on pre-treatment and early follow-up multi-parametric MRI in LARC patients receiving CRT could extract comprehensive quantitative information to predict final pathologic response.

Automatic Breast and Fibroglandular Tissue Segmentation in Breast MRI Using Deep Learning by a Fully-Convolutional Residual Neural Network U-Net.

RATIONALE AND OBJECTIVES: Breast segmentation using the U-net architecture was implemented and tested in independent validation datasets to quantify fibroglandular tissue volume in breast MRI. MATERIALS AND METHODS: Two datasets were used. The training set was MRI of 286 patients with unilateral breast cancer. The segmentation was done on the contralateral normal breasts. The ground truth for the breast and fibroglandular tissue (FGT) was obtained by using a template-based segmentation method. The U-net deep learning algorithm was implemented to analyze the training set, and the final model was obtained using 10-fold cross-validation. The independent validation set was MRI of 28 normal volunteers acquired using four different MR scanners. Dice Similarity Coefficient (DSC), voxel-based accuracy, and Pearson's correlation were used to evaluate the performance. RESULTS: For the 10-fold cross-validation in the initial training set of 286 patients, the DSC range was 0.83-0.98 (mean 0.95 ± 0.02) for breast and 0.73-0.97 (mean 0.91 ± 0.03) for FGT; and the accuracy range was 0.92-0.99 (mean 0.98 ± 0.01) for breast and 0.87-0.99 (mean 0.97 ± 0.01) for FGT. For the entire 224 testing breasts of the 28 normal volunteers in the validation datasets, the mean DSC was 0.86 ± 0.05 for breast, 0.83 ± 0.06 for FGT; and the mean accuracy was 0.94 ± 0.03 for breast and 0.93 ± 0.04 for FGT. The testing results for MRI acquired using four different scanners were comparable. CONCLUSION: Deep learning based on the U-net algorithm can achieve accurate segmentation results for the breast and FGT on MRI. It may provide a reliable and efficient method to process large number of MR images for quantitative analysis of breast density.

Temperature-modulated fluorescence tomography based on both concentration and lifetime contrast.

It is challenging to image fluorescence objects with high spatial resolution in a highly scattering medium. Recently reported temperature-sensitive indocyanine green-loaded pluronic nanocapsules can potentially alleviate this problem. Here we demonstrate a frequency-domain temperature-modulated fluorescence tomography system that could acquire images at high intensity-focused ultrasound resolution with use of these nanocapsules. The system is experimentally verified with a phantom study, where a 3-mm fluorescence object embedded 2 cm deep in a turbid medium is successfully recovered based on both intensity and lifetime contrast.

Temperature-modulated fluorescence tomography in a turbid media.

High scattering in biological tissues makes fluorescence tomography inverse problem very challenging in thick medium. We describe an approach termed "temperature-modulated fluorescence tomography" that can acquire fluorescence images at focused ultrasound resolution. By utilizing recently emerged temperature sensitive fluorescence contrast agents, this technique provides fluorescence images with high resolution prior to any reconstruction process. We demonstrate that this technique is well suited to resolve small fluorescence targets located several centimeters deep in tissue.