Delineation Of Lesions Research Articles

A deep supervised cross-attention strategy for ischemic stroke segmentation in MRI studies

The key component of stroke diagnosis is the localization and delineation of brain lesions, especially from MRI studies. Nonetheless, this manual delineation is time-consuming and biased by expert opinion. The main purpose of this study is to introduce an autoencoder architecture that effectively integrates cross-attention mechanisms, together with hierarchical deep supervision to delineate lesions under scenarios of remarked unbalance tissue classes, challenging geometry of the shape, and a variable textural representation. This work introduces a cross-attention deep autoencoder that focuses on the lesion shape through a set of convolutional saliency maps, forcing skip connections to preserve the morphology of affected tissue. Moreover, a deep supervision training scheme was herein adapted to induce the learning of hierarchical lesion details. Besides, a special weighted loss function remarks lesion tissue, alleviating the negative impact of class imbalance. The proposed approach was validated on the public ISLES2017 dataset outperforming state-of-the-art results, achieving a dice score of 0.36 and a precision of 0.42. Deeply supervised cross-attention autoencoders, trained to pay more attention to lesion tissue, are better at estimating ischemic lesions in MRI studies. The best architectural configuration was achieved by integrating ADC, TTP and Tmax sequences. The contribution of deeply supervised cross-attention autoencoders allows better support the discrimination between healthy and lesion regions, which in consequence results in favorable prognosis and follow-up of patients.

Diagnostic value of dynamic 18F-FDG PET/CT imaging in non-small cell lung cancer and FDG hypermetabolic lymph nodes.

We aimed to investigate the diagnostic value of dynamic 2-deoxy-2-[18F]-fluoro-D-glucose positron emission tomography-computed tomography in non-small cell lung cancer and fluoro-D-glucose hypermetabolic lymph nodes. Patients who made an active appointment for positron emission tomography-computed tomography were randomly enrolled by referring to previous imaging data and clinical information. Finally, 34 histopathologically confirmed non-small cell lung cancers (18 adenocarcinoma and 16 squamous cell carcinoma cases) were prospectively studied using dynamic and static 2-deoxy-2-[18F]-fluoro-D-glucose positron emission tomography-computed tomography imaging (the diagnostic study has not yet been registered on a clinical trial platform). In dynamic positron emission tomography images, a volume of interest, defined by the thoracic aorta, was selected for estimating the arterial input function. Patlak and irreversible two-tissue compartment model analyses were performed based on the pixel points to obtain first-order characteristic kinetic parameters for each lesion and hypermetabolic lymph node. The first-order characteristic kinetic parameters were obtained based on the basic data of dynamic positron emission tomography images in the corresponding model and the lesion delineation of region-of-interest based on computed tomography images, such as V_Median (the median gray intensity of V), k3_Entropy, VB_Entropy, K1_Uniformity, and ki_Uniformity. The first-order characteristic kinetic parameters were also modeled by logistic regression for the differential diagnosis of non-small cell lung cancer and hypermetabolic lymph nodes. Maximum and mean standard uptake values (SUVmax and SUVmean, respectively) were obtained from static positron emission tomography images. The diagnostic efficacy of the parameters was evaluated using the receiver operating characteristic curve and the DeLong test. There was a significant difference in the V_Median values of adenocarcinoma and squamous cell carcinoma. The regression models for K1, k3, and V provided good predictions of adenocarcinoma and squamous cell carcinoma typology. Significant differences were observed in k3_Entropy, VB_Entropy, K1_Uniformity, and ki_Uniformity between benign and malignant lymph nodes. The regression model of Ki, VB, and k3 could make a good prediction for identifying benign and malignant lymph nodes. Dynamic 2-deoxy-2-[18F]-fluoro-D-glucose positron emission tomography-computed tomography imaging showed high diagnostic value in the staging of non-small cell lung cancer and fluoro-D-glucose hypermetabolic lymph nodes, and can be of great use in non-small cell lung cancer lymph node staging and surgical decision-making.

Deep learning-based reconstruction can improve the image quality of low radiation dose head CT.

To evaluate the image quality of deep learning-based reconstruction (DLR), model-based (MBIR), and hybrid iterative reconstruction (HIR) algorithms for lower-dose (LD) unenhanced head CT and compare it with those of standard-dose (STD) HIR images. This retrospective study included 114 patients who underwent unenhanced head CT using the STD (n = 57) or LD (n = 57) protocol on a 320-row CT. STD images were reconstructed with HIR; LD images were reconstructed with HIR (LD-HIR), MBIR (LD-MBIR), and DLR (LD-DLR). The image noise, gray and white matter (GM-WM) contrast, and contrast-to-noise ratio (CNR) at the basal ganglia and posterior fossa levels were quantified. The noise magnitude, noise texture, GM-WM contrast, image sharpness, streak artifact, and subjective acceptability were independently scored by three radiologists (1 = worst, 5 = best). The lesion conspicuity of LD-HIR, LD-MBIR, and LD-DLR was ranked through side-by-side assessments (1 = worst, 3 = best). Reconstruction times of three algorithms were measured. The effective dose of LD was 25% lower than that of STD. Lower image noise, higher GM-WM contrast, and higher CNR were observed in LD-DLR and LD-MBIR than those in STD (all, p ≤ 0.035). Compared with STD, the noise texture, image sharpness, and subjective acceptability were inferior for LD-MBIR and superior for LD-DLR (all, p < 0.001). The lesion conspicuity of LD-DLR (2.9 ± 0.2) was higher than that of HIR (1.2 ± 0.3) and MBIR (1.8 ± 0.4) (all, p < 0.001). Reconstruction times of HIR, MBIR, and DLR were 11 ± 1, 319 ± 17, and 24 ± 1s, respectively. DLR can enhance the image quality of head CT while preserving low radiation dose level and short reconstruction time. •For unenhanced head CT, DLR reduced the image noise and improved the GM-WM contrast and lesion delineation without sacrificing the natural noise texture and image sharpness relative to HIR. •The subjective and objective image quality of DLR was better than that of HIR even at 25% reduced dose without considerably increasing the image reconstruction times (24s vs. 11s). •Despite the strong noise reduction and improved GM-WM contrast performance, MBIR degraded the noise texture, sharpness, and subjective acceptance with prolonged reconstruction times relative to HIR, potentially hampering its feasibility.

The elusive metric of lesion load.

One of the widely used metrics in lesion-symptom mapping is lesion load that codes the amount of damage to a given brain region of interest. Lesion load aims to reduce the complex 3D lesion information into a feature that can reflect both site of damage, defined by the location of the region of interest, and size of damage within that region of interest. Basically, the process of estimation of lesion load converts a voxel-based lesion map into a region-based lesion map, with regions defined as atlas-based or data-driven spatial patterns. Here, after examining current definitions of lesion load, four methodological issues are discussed: (1) lesion load is agnostic to the location of damage within the region of interest, and it disregards damage outside the region of interest, (2) lesion load estimates are prone to errors introduced by the uncertainty in lesion delineation, spatial warping of the lesion/region, and binarization of the lesion/region, (3) lesion load calculation depends on brain parcellation selection, and (4) lesion load does not necessarily reflect a white matter disconnection. Overall, lesion load, when calculated in a robust way, can serve as a clinically-useful feature for explaining and predicting post-stroke outcome and recovery.

Abstract P3-04-05: Artificial Intelligence Detects, Classifies, and Describes Lesions in Clinical Breast Ultrasound Images

Abstract Purpose Many low-middle income countries (LMIC) suffer from chronic shortages of resources that inhibit the implementation of effective breast cancer screening programs. Advanced breast cancer rates in the U.S. Affiliated Pacific Islands substantially exceed that of the United States. We propose the use of portable breast ultrasound coupled with artificial intelligence (AI) algorithms to assist non-radiologist field personnel in real-time field lesion detection, classification, and biopsy, as well as determination of breast density for risk assessment. In this study, we examine the ability of an AI algorithm to detect and describe breast cancer lesions in clinically-acquired breast ultrasound images in 40,000 women participating in a Hawaii screening program. Materials and Methods The Hawaii and Pacific Islands Mammography Registry (HIPIMR) collects breast health questionnaires and breast imaging (mammography, ultrasound, and MRI) from participating centers in Hawaii and the Pacific and links this information to the Hawaii Tumor Registry for cancer findings. From the women with either screening or diagnostic B-mode breast ultrasound exams, we selected 3 negative cases (no cancer) for every positive case matched by age, excluding Doppler and elastography images. The blinded images were read by the study radiologist to delineate all lesions and describe in terms of the BI-RADS lexicon. These images were split by woman into training (70%), validation and hyperparameter selection (20%) and testing (20%) subsets. An AI model was fine-tuned for lesion and BI-RADS category classification from a Detectron2 framework [1] pre-trained on the COCO Instance Segmentation Dataset [2]. Model performance was evaluated by computation of precision and sensitivity percentages, as well as Area under the Receiver Operator Curve (AUROC). Detections were considered positive if they overlapped a ground truth lesion delineation by at least 50% (Intersection over Union = 0.5), and a maximum of 4 detections were generated for each image. Timing experiments were run on a GPU-enabled (Nvidia Tesla V100) machine on unbatched images. Results Over the 10-year observation period, we identified 5,214 women with US images meeting our criterion. Of these, 111 were diagnosed with malignant breast cancer and 333 were selected as non-cases for a total of 444 women. These 444 women had a total of 4,623 ultrasound images with 2,028 benign and 1,431 malignant lesions identified by the study radiologist. For cancerous lesions, the AI algorithm had 8% precision at a sensitivity of 90% on the testing set. For benign lesions, a sensitivity of 90% resulted in 5% precision on the testing set. The AUROC for bounding box detections of cancerous lesions was 0.90. The AUROC for bounding box detections of benign lesions was 0.87. The model made predictions at a rate of 25 frames/second time (38.7 milliseconds per image). Conclusion Detection, segmentation, and cancer classification of breast lesions are possible in clinically-acquired ultrasound images using AI. Based on our timing experiments, the model is capable of detecting and classifying lesions in real-time during ultrasound capture. Model performance is expected to improve as more data becomes available for training. Future work would involve further fine-tuning of the model on portable breast ultrasound images and increasing model evaluation speed in order to assess utility in low-resource populations [1] Wu Y, Kirillov A, Massa F, Lo W-Y, Girshick R. Detectron2. https://github.com/facebookresearch/detectron2. [2] Lin T-Y, Maire M, Belongie S, et al. Microsoft COCO: Common Objects in Context. Computer Vision – ECCV 2014. Springer International Publishing; 2014:740-755. Citation Format: Arianna Bunnell, Dustin Valdez, Thomas Wolfgruber, Aleen Altamirano, Brenda Hernandez, Peter Sadowski, John Shepherd. Artificial Intelligence Detects, Classifies, and Describes Lesions in Clinical Breast Ultrasound Images [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P3-04-05.

Repeatability of metabolic tumor burden and lesion glycolysis between clinical readers.

The Metabolic Tumor Volume (MTV) and Tumor Lesion Glycolysis (TLG) has been shown to be independent prognostic predictors for clinical outcome in Diffuse Large B-cell Lymphoma (DLBCL). However, definitions of these measurements have not been standardized, leading to many sources of variation, operator evaluation continues to be one major source. In this study, we propose a reader reproducibility study to evaluate computation of TMV (& TLG) metrics based on differences in lesion delineation. In the first approach, reader manually corrected regional boundaries after automated detection performed across the lesions in a body scan (Reader M using a manual process, or manual). The other reader used a semi-automated method of lesion identification, without any boundary modification (Reader A using a semi- automated process, or auto). Parameters for active lesion were kept the same, derived from standard uptake values (SUVs) over a 41% threshold. We systematically contrasted MTV & TLG differences between expert readers (Reader M & A). We find that MTVs computed by Readers M and A were both concordant between them (concordant correlation coefficient of 0.96) and independently prognostic with a P-value of 0.0001 and 0.0002 respectively for overall survival after treatment. Additionally, we find TLG for these reader approaches showed concordance (CCC of 0.96) and was prognostic for over -all survival (p ≤ 0.0001 for both). In conclusion, the semi-automated approach (Reader A) provides acceptable quantification & prognosis of tumor burden (MTV) and TLG in comparison to expert reader assisted measurement (Reader M) on PET/CT scans.

Application of artificial intelligence to stereotactic radiosurgery for intracranial lesions: detection, segmentation, and outcome prediction.

Rapid evolution of artificial intelligence (AI) prompted its wide application in healthcare systems. Stereotactic radiosurgery served as a good candidate for AI model development and achieved encouraging result in recent years. This article aimed at demonstrating current AI application in radiosurgery. Literatures published in PubMed during 2010-2022, discussing AI application in stereotactic radiosurgery were reviewed. AI algorithms, especially machine learning/deep learning models, have been administered to different aspect of stereotactic radiosurgery. Spontaneous tumor detection and automated lesion delineation or segmentation were two of the promising application, which could be further extended to longitudinal treatment follow-up. Outcome prediction utilized machine learning algorithms with radiomic-based analysis was another well-established application. Stereotactic radiosurgery has taken a lead role in AI development. Current achievement, limitation, and further investigation was summarized in this article.

Predicting the Onset of Ischemic Stroke With Fast High-Resolution 3D MR Spectroscopic Imaging.

Neurometabolite concentrations provide a direct index of infarction progression in stroke. However, their relationship with stroke onset time remains unclear. To assess the temporal dynamics of N-acetylaspartate (NAA), creatine, choline, and lactate and estimate their value in predicting early (<6 hours) vs. late (6-24hours) hyperacute stroke groups. Cross-sectional cohort. A total of 73 ischemic stroke patients scanned at 1.8-302.5 hours after symptom onset, including 25 patients with follow-up scans. A 3 T/magnetization-prepared rapid acquisition gradient echo sequence for anatomical imaging, diffusion-weighted imaging and fluid-attenuated inversion recovery imaging for lesion delineation, and 3D MR spectroscopic imaging (MRSI) for neurometabolic mapping. Patients were divided into hyperacute (0-24hours), acute (24hours to 1 week), and subacute (1-2 weeks) groups, and into early (<6 hours) and late (6-24hours) hyperacute groups. Bayesian logistic regression was used to compare classification performance between early and late hyperacute groups by using different combinations of neurometabolites as inputs. Linear mixed effects modeling was applied for group-wise comparisons between NAA, creatine, choline, and lactate. Pearson's correlation analysis was used for neurometabolites vs. time. P < 0.05 was considered statistically significant. Lesional NAA and creatine were significantly lower in subacute than in acute stroke. The main effects of time were shown on NAA (F=14.321) and creatine (F= 12.261). NAA was significantly lower in late than early hyperacute patients, and was inversely related to time from symptom onset across both groups (r=-0.440). The decrease of NAA and increase of lactate were correlated with lesion volume (NAA: r=-0.472; lactate: r=0.366) in hyperacute stroke. Discrimination was improved by combining NAA, creatine, and choline signals (area under the curve [AUC]=0.90). High-resolution 3D MRSI effectively assessed the neurometabolite changes and discriminated early and late hyperacute stroke lesions. 1. Stage 2.

BUS-Net: Breast Tumour Detection Network for Ultrasound Images Using Bi-directional ConvLSTM and Dense Residual Connections.

Breast ultrasound (BUS) imaging has become one of the key imaging modalities for medical image diagnosis and prognosis. However, the manual process of lesion delineation from ultrasound images can incur various challenges concerning variable shape, size, intensity, curvature, or other medical priors of the lesion in the image. Therefore, computer-aided diagnostic (CADx) techniques incorporating deep learning-based neural networks are automatically used to segment the lesion from BUS images. This paper proposes an encoder-decoder-based architecture to recognize and accurately segment the lesion from two-dimensional BUS images. The architecture is utilized with the residual connection in both encoder and decoder paths; bi-directional ConvLSTM (BConvLSTM) units in the decoder extract the minute and detailed region of interest (ROI) information. BConvLSTM units and residual blocks help the network weigh ROI information more than the similar background region. Two public BUS image datasets, one with 163 images and the other with 42 images, are used. The proposed model is trained with the augmented images (ten forms) of dataset one (with 163 images), and test results are produced on the second dataset and the testing set of the first dataset-the segmentation performance yielding comparable results with the state-of-the-art segmentation methodologies. Similarly, the visual results show that the proposed approach for BUS image segmentation can accurately identify lesion contours and can potentially be applied for similar and larger datasets.

SURG-37. NEURO-ONCOLOGICAL AUGMENTED REALITY: A MORE INTUITIVE APPROACH TO RESECTION PLANNING

Abstract BACKGROUND When preparing for the resection of an intracranial lesion, its borders and optimal approach are often determined using neuronavigation with a tracked pointer. This can sometimes prove challenging, especially for deep-seated lesions. Augmented reality (AR) can simplify and improve this step by directly displaying the lesion on the patient's skin. METHODS A proprietary inside-out infrared tracking solution was developed, allowing for heads-up displaying AR scenes on the Microsoft HoloLens II without the need for an external tracking camera or computer. We included twenty patients with an intracerebral lesion planned for resection. After semi-automatic hologram-to-patient registration, different participants marked the lesion outlines on the patient’s skin, consecutively aided by the Brainlab neuronavigation system and the HoloLens. Each registration on both systems provided a registration transform that was compared for accuracy and consistency. Participant performance was quantified in terms of duration and accuracy for both patient registration and lesion delineation, and compared to expert performance. RESULTS When using AR both registration and delineation were significantly faster than with conventional neuronavigation (p = 0.02 and p &amp;lt; 0.001, respectively, and p &amp;lt; 0.001 for the total duration), taking 79.23 ± 17.48 and 39.58 ± 39.10 seconds while neuronavigation required 96.61 ± 24.54 and 90.80 ± 44.09 seconds. AR had a registration offset of 3.3mm and 3.4°, and was more consistent compared to neuronavigation. AR facilitated more accurate and detailed lesion delineation, while neuronavigation often overestimated lesion size. CONCLUSION Augmented reality provides a faster and more accurate alternative for resection planning. Lesion delineation was more intuitive while retaining high accuracy. Future research should focus on further intraoperative implementations.

Myeloma bone disease imaging on a 1st-generation clinical photon-counting detector CT vs. 2nd-generation dual-source dual-energy CT

ObjectiveSubjective and objective image quality comparison of bone microstructure and disease-related abnormalities in multiple myeloma patients using a 1st-generation dual-source photon-counting detector CT(DS-PCD-CT) and a 2nd-generation dual-source dual-energy (energy-integrating detector) CT (DS-EID-CT).MethodsFifty multiple myeloma patients (mean age 67.7 ± 10.9 years,16 females) were prospectively enrolled. Unenhanced whole-body CTs were clinically indicated and performed on DS-EID-CT and DS-PCD-CT (median time difference: 12 months). DS-PCD-CT was performed in Quantumplus UHR mode and DS-EID-CT was performed using dual-energy mode. DS-PCD-CT kernel was set at Br64 with Quantum iterative reconstruction strength Q1; for DS-EID-CT a comparable I70f kernel with SAFIRE iterative reconstruction strength 1 was used. Two independent radiologists assessed image quality subjectively using a 5-point Likert scale considering delineation and sharpness of trabecular bone and lytic bone lesions in the spine and pelvic bones. Additionally, ImageJ was used for quantification of bony septa inside the cancellous bone and through or the edges of osteolysis.ResultsOverall quality as well as detectability and sharpness in the delineation of lytic bone lesions were superior for DS-PCD-CT compared with DS-EID-CT (p < 0.0001). The inter-reader agreement for subjective image quality readings showed excellent consistency(α = 94.2–98.8). CTDI and DLP mean values for DS-PCD-CT and DS-EID-CT were 1107.4 ± 247.6 mGy*cm and 8.2 ± 1.8 mGy vs. 1344.3 ± 204.6 mGy*cm and 10.1 ± 1.9 mGy. The quantitative metric for bone microstructure in the femoral head showed significantly better visualization of trabeculae in DS-PCD-CT compared with DS-EID-CT (p < 0.0001). Quantitative analyses of edge sharpness of osteolysis showed significant steeper edges for DS-PCD-CT (p < 0.0001).ConclusionDS-PCD-CT significantly improves spatial resolution of bony microstructure and lytic bone lesions compared to DS-EID-CT.Key Points• Application of photon-counting detector CT is superior to dual-source dual-energy integrating detector in clinical workup of multiple myeloma patients.• Compared to energy integrating detectors, photon-counting detectors significantly increase the spatial resolution of bone microstructure including disease-related lytic bone lesions in patients with multiple myeloma.

New multiple sclerosis lesion segmentation and detection using pre-activation U-Net.

Automated segmentation of new multiple sclerosis (MS) lesions in 3D MRI data is an essential prerequisite for monitoring and quantifying MS progression. Manual delineation of such lesions is time-consuming and expensive, especially because raters need to deal with 3D images and several modalities. In this paper, we propose Pre-U-Net, a 3D encoder-decoder architecture with pre-activation residual blocks, for the segmentation and detection of new MS lesions. Due to the limited training set and the class imbalance problem, we apply intensive data augmentation and use deep supervision to train our models effectively. Following the same U-shaped architecture but different blocks, Pre-U-Net outperforms U-Net and Res-U-Net on the MSSEG-2 dataset, achieving a Dice score of 40.3% on new lesion segmentation and an F1 score of 48.1% on new lesion detection. The codes and trained models are publicly available at https://github.com/pashtari/xunet.

Whole-tumor apparent diffusion coefficient (ADC) analyses of breast lesions based on simultaneous multi-slice readout-segmented echo-planar diffusion-weighted imaging

ObjectiveWe aimed to evaluate the effectiveness of simultaneous multi-slice readout-segmented echo-planar diffusion-weighted imaging (SMS rs-EPI DWI), compared with readout-segmented echo-planar diffusion-weighted imaging (NOSMS rs-EPI DWI), in discriminating between benign and malignant breast lesions. Materials and methodsThis retrospective study evaluated breast lesions from 185 consecutive patients who had undergone preoperative breast MRI. The NOSMS rs-EPI DWI and the prototype SMS rs-EPI DWI sequences were performed on a 1.5-T MR scanner. Two independent radiologists evaluated the image quality of the two sequences using a 5-point scale (1 = poor, 5 = excellent) and then assessed the scores through visual grading characteristics analysis (VGC). The values of signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and whole-tumor-based histogram parameter (ADCmedian) were compared between the two sequences using the Wilcoxon test. Then ROC curves were used for statistical analysis. ResultsThe visual assessment showed that the SMS rs-EPI yielded superior overall image quality and lesion delineation than the NOSMS rs-EPI (AUCVGC = 0.980 and 0.984, respectively; all P < 0.001). There were no significant differences in relevant artifacts between the two sequences (AUCVGC = 0.531, P = 0.462). The SNR values for SMS rs-EPI DWI were significantly lower than for NOSMS rs-EPI DWI (P = 0.019) while there was no significant difference in CNR values between the two sequences (P = 0.955). In addition, evaluation of the diagnostic performance demonstrated that the difference in ADCmedian values for both DWI sequences between the malignant and benign lesions was statistically significant (P < 0.001). In contrast, the AUC for ADCmedian was higher with SMS rs-EPI than NOSMS rs-EPI (0.879 for SMS vs. 0.839 for NOSMS, P = 0.025). ConclusionThe SMS technique could further improve the image quality and the diagnostic performance of rs-EPI DWI in a comparable time.

Impact of SSTR PET on Inter-Observer Variability of Target Delineation of Meningioma and the Possibility of Using Threshold-Based Segmentations in Radiation Oncology.

Simple SummaryDifferences in tumor segmentations between radiation oncologists is one of the largest sources of uncertainty in radiation therapy planning. This study investigated the influence of additional functional information from somatostatin receptor PET imaging on the inter observer variability in the delineation of meningioma. Further, this study assessed the usability of a simple thresholding approach for lesion delineation. It could be shown, that additional PET information was able to significantly reduce the inter observer variability. The threshold based delineation approach required a relatively low threshold value and showed only moderate agreement with the radiation oncologists.Aim: The aim of this study was to assess the effects of including somatostatin receptor agonist (SSTR) PET imaging in meningioma radiotherapy planning by means of changes in inter-observer variability (IOV). Further, the possibility of using threshold-based delineation approaches for semiautomatic tumor volume definition was assessed. Patients and Methods: Sixteen patients with meningioma undergoing fractionated radiotherapy were delineated by five radiation oncologists. IOV was calculated by comparing each delineation to a consensus delineation, based on the simultaneous truth and performance level estimation (STAPLE) algorithm. The consensus delineation was used to adapt a threshold-based delineation, based on a maximization of the mean Dice coefficient. To test the threshold-based approach, seven patients with SSTR-positive meningioma were additionally evaluated as a validation group. Results: The average Dice coefficients for delineations based on MRI alone was 0.84 ± 0.12. For delineation based on MRI + PET, a significantly higher dice coefficient of 0.87 ± 0.08 was found (p < 0.001). The Hausdorff distance decreased from 10.96 ± 11.98 mm to 8.83 ± 12.21 mm (p < 0.001) when adding PET for the lesion delineation. The best threshold value for a threshold-based delineation was found to be 14.0% of the SUVmax, with an average Dice coefficient of 0.50 ± 0.19 compared to the consensus delineation. In the validation cohort, a Dice coefficient of 0.56 ± 0.29 and a Hausdorff coefficient of 27.15 ± 21.54 mm were found for the threshold-based approach. Conclusions: SSTR-PET added to standard imaging with CT and MRI reduces the IOV in radiotherapy planning for patients with meningioma. When using a threshold-based approach for PET-based delineation of meningioma, a relatively low threshold of 14.0% of the SUVmax was found to provide the best agreement with a consensus delineation.

P07.02.B Neuro-oncological augmented reality planning (NOA-p)

Abstract Background When preparing for the resection of an intracranial lesion, neuronavigation with a tracked pointer is most often used to determine lesion borders and the optimal approach. This can sometimes prove challenging, especially for deep-seated lesions. Augmented Reality (AR), directly displaying the lesion on the patient’s skin, can simplify and improve this step. Material and Methods We developed a system for inside-out infrared tracking that does not require an external tracking camera or external computer and allows for heads-up displaying an AR scene on the Microsoft HoloLens II. Twenty patients planned for the resection of an intracerebral lesion were included in our study. After patient registration, the lesion outlines were marked on the patient’s skin by different participants, consecutively using the Brainlab neuronavigation system and the HoloLens. Each registration on both systems provided a registration transform that was compared for accuracy and consistency. The performance of the participants was measured in terms of duration and accuracy and compared to expert registration and delineation. Results Both registration and delineation were significantly faster with AR (p=0.02 and p&amp;lt;0.001, respectively, and p&amp;lt;0.001 for the total duration), taking 79.23±17.48 and 39.58±39.10 seconds while neuronavigation required 96.61±24.54 and 90.80±44.09 seconds. AR had a registration offset of 3.3mm and 3.4°, and was more consistent compared to neuronavigation. AR facilitated more accurate and detailed lesion delineation, while neuronavigation often overestimated lesion size. Conclusion Augmented reality provides a faster and more accurate alternative for resection planning. Lesion delineation is more intuitive while remaining high in accuracy. Future research should focus on further intraoperative implementations.

Deep learning-based virtual noncalcium imaging in multiple myeloma using dual-energy CT.

Dual-energy CT with virtual noncalcium (VNCa) images allows the evaluation of focal intramedullary bone marrow involvement in patients with multiple myeloma. However, current commercial VNCa techniques suffer from excessive image noise and artifacts due to material decomposition used in synthesizing VNCa images. In this work, we aim to improve VNCa image quality for the assessment of focal multiple myeloma, using an Artificial intelligence based Generalizable Algorithm for mulTi-Energy CT (AGATE) method. AGATE method used a custom dual-task convolutional neural network (CNN) that concurrently carries out material classification and quantification. The material classification task provided an auxiliary regularization to the material quantification task. CNN parameters were optimized using custom loss functions that involved cross-entropy, physics-informed constraints, structural redundancy in spectral and material images, and texture information in spectral images. For training data, CT phantoms (diameters 30 to 45cm)withtissue-mimicking inserts were scanned on a third generation dual-source CT system. Scans were performed at routine dose and half of the routine dose. Small image patches (i.e., 40 × 40 pixels) of tissue-mimicking inserts with known basis material densities were extracted for training samples. Numerically simulated insert materials with various shapes increased diversity of training samples. Generalizability of AGATE was evaluated using CT images from phantoms and patients. In phantoms, material decomposition accuracy was estimated using mean-absolute-percent-error (MAPE), using physical inserts that were not used during the training. Noise power spectrum (NPS) and modulation transfer function (MTF) were compared across phantom sizes and radiation dose levels. Five patients with multiple myeloma underwent dual-energy CT, with VNCa images generated using a commercial method and AGATE. Two fellowship-trained musculoskeletal radiologists reviewed the VNCa images (commercial and AGATE) side-by-side using a dual-monitor display, blinded to VNCa type, rating the image quality for focal multiple myeloma lesion visualization using a 5-level Likert comparison scale (-2=worse visualization and diagnostic confidence, -1=worse visualization but equivalent diagnostic confidence, 0=equivalent visualization and diagnostic confidence, 1=improved visualization but equivalent diagnostic confidence, 2=improved visualization and diagnostic confidence). A post hoc assignment of comparison ratings was performed to rank AGATE images in comparison to commercial ones. AGATE demonstrated consistent material quantification accuracy across phantom sizes and radiation dose levels, with MAPE ranging from 0.7% to 4.4% across all testing materials. Compared to commercial VNCa images, the AGATE-synthesized VNCa images yielded considerably lower image noise (50-77% noise reduction) without compromising noise texture or spatial resolution across different phantom sizes and two radiation doses. AGATE VNCa images had markedly reduced area under NPS curves and maintained NPS peak frequency (0.7 lp/cm to 1.0 lp/cm), with similar MTF curves (50% MTF at 3.0 lp/cm). In patients, AGATE demonstrated reduced image noise and artifacts with improved delineation of focal multiple myeloma lesions (all readers comparison scores indicating improved overall diagnostic image quality [scores 1 or 2]). AGATE demonstrated reduced noise and artifacts in VNCa images and ability to improve visualization of bone marrow lesions for assessing multiple myeloma.

Functional imaging and targeted drug delivery in mice and patient tumors with a cell nucleolus-localizing and tumor-targeting peptide

Tumor-targeting peptides have profound clinical implications in early detection and delineation of microscopic lesions for surgical resection, and also delivery of therapeutics with reduced systemic toxicity. Here, we demonstrate that a peptide (RS), evolved from a previously reported hepatocellular carcinoma (HCC)-targeting peptide P47, enables improved HCC micrometastasis discrimination and delineation from noncancerous tissues in murine orthotopic mice and patient biopsies, with up to 21-fold contrast. Importantly, RS targets non-small cell lung (NSCLC) and colon cancers in mice and patient biopsies, with higher selectivity for highly proliferative tumor nodules. Moreover, RS localizes to cell nucleoli of HCC, NSCLC, breast, colon and cervical cancer cells and induces nucleolar stress when conjugated with chemotherapeutic Oxaliplatin (OXA) (RS-OXA), demonstrating both cellular and subcellular targeting. RS-delivered OXA elicits significant tumor retardation in orthotopic HCC mice with markedly reduced systemic toxicity compared to OXA alone. Injection of fluorescence-labeled RS enables dynamic visualization of tumor growth in RS-OXA-treated subcutaneous HCC mice. Our study demonstrates that RS targets a spectrum of tumors and localizes to cell nucleolus, thus enabling functional imaging and targeted delivery of OXA in HCC mice, and consequently provides a versatile tool for tumor imaging and targeted therapeutics.

Field validation of deep learning based Point-of-Care device for early detection of oral malignant and potentially malignant disorders

Early detection of oral cancer in low-resource settings necessitates a Point-of-Care screening tool that empowers Frontline-Health-Workers (FHW). This study was conducted to validate the accuracy of Convolutional-Neural-Network (CNN) enabled m(mobile)-Health device deployed with FHWs for delineation of suspicious oral lesions (malignant/potentially-malignant disorders). The effectiveness of the device was tested in tertiary-care hospitals and low-resource settings in India. The subjects were screened independently, either by FHWs alone or along with specialists. All the subjects were also remotely evaluated by oral cancer specialist/s. The program screened 5025 subjects (Images: 32,128) with 95% (n = 4728) having telediagnosis. Among the 16% (n = 752) assessed by onsite specialists, 20% (n = 102) underwent biopsy. Simple and complex CNN were integrated into the mobile phone and cloud respectively. The onsite specialist diagnosis showed a high sensitivity (94%), when compared to histology, while telediagnosis showed high accuracy in comparison with onsite specialists (sensitivity: 95%; specificity: 84%). FHWs, however, when compared with telediagnosis, identified suspicious lesions with less sensitivity (60%). Phone integrated, CNN (MobileNet) accurately delineated lesions (n = 1416; sensitivity: 82%) and Cloud-based CNN (VGG19) had higher accuracy (sensitivity: 87%) with tele-diagnosis as reference standard. The results of the study suggest that an automated mHealth-enabled, dual-image system is a useful triaging tool and empowers FHWs for oral cancer screening in low-resource settings.

Evaluation of the ability of the Brainlab Elements Cranial Distortion Correction algorithm to correct clinically relevant MRI distortions for cranial SRT.

Cranial stereotactic radiotherapy (SRT) requires highly accurate lesion delineation. However, MRI can have significant inherent geometric distortions. We investigated how well the Elements Cranial Distortion Correction algorithm of Brainlab (Munich, Germany) corrects the distortions in MR image-sets of aphantom and patients. Anon-distorted reference computed tomography image-set of aCIRS Model 603-GS (CIRS, Norfolk, VA, USA) phantom was acquired. Three-dimensional T1-weighted images were acquired with five MRI scanners and reconstructed with vendor-derived distortion correction. Some were reconstructed without correction to generate heavily distorted image-sets. All MR image-sets were corrected with the Brainlab algorithm relative to the computed tomography acquisition. CIRS Distortion Check software measured the distortion in each image-set. For all uncorrected and corrected image-sets, the control points that exceeded the 0.5-mm clinically relevant distortion threshold and the distortion maximum, mean, and standard deviation were recorded. Empirical cumulative distribution functions (eCDF) were plotted. Intraclass correlation coefficient (ICC) was calculated. The algorithm was evaluated with 10brain metastases using Dice similarity coefficients (DSC). The algorithm significantly reduced mean and standard deviation distortion in all image-sets. It reduced the maximum distortion in the heavily distorted image-sets from 2.072 to 1.059 mm and the control points with > 0.5-mm distortion fell from 50.2% to 4.0%. Before and especially after correction, the eCDFs of the four repeats were visually similar. ICC was 0.812 (excellent-good agreement). The algorithm increased the DSCs for all patients and image-sets. The Brainlab algorithm significantly and reproducibly ameliorated MRI distortion, even with heavily distorted images. Thus, it increases the accuracy of cranial SRT lesion delineation. After further testing, this tool may be suitable for SRT of small lesions.

Recent advances in the longitudinal segmentation of multiple sclerosis lesions on magnetic resonance imaging: a review.

Multiple sclerosis (MS) is a chronic autoimmune disease characterized by demyelinating lesions that are often visible on magnetic resonance imaging (MRI). Segmentation of these lesions can provide imaging biomarkers of disease burden that can help monitor disease progression and the imaging response to treatment. Manual delineation of MRI lesions is tedious and prone to subjective bias, while automated lesion segmentation methods offer objectivity and speed, the latter being particularly important when analysing large datasets. Lesion segmentation can be broadly categorised into two groups: cross-sectional methods, which use imaging data acquired at a single time-point to characterise MRI lesions; and longitudinal methods, which use imaging data from the same subject acquired at two or more different time-points to characterise lesions over time. The main objective of longitudinal segmentation approaches is to more accurately detect the presence of new MS lesions and the growth or remission of existing lesions, which may be effective biomarkers of disease progression and treatment response. This paper reviews articles on longitudinal MS lesion segmentation methods published over the past 10years. These are divided into traditional machine learning methods and deep learning techniques. PubMed articles using longitudinal information and comparing fully automatic two time point segmentations in any step of the process were selected. Nineteen articles were reviewed. There is an increasing number of deep learning techniques for longitudinal MS lesion segmentation that are promising to help better understand disease progression.