Tumor spheroid growth assays are used to evaluate the potential of cancer therapies in vitro. During such experiments, extensive microscopic image series are generated, which are commonly analyzed using threshold-based delineations. However, due to treatment-induced morphological changes of the spheroids, very time-consuming manual corrections are often required. The goal of our work was the development of an AI-based method for accurate and automated delineation of spheroid growth assays, ultimately reducing the reliance on manual delineation and corrections. Spheroids were grown from mouse pheochromocytoma (MPC) cells and subjected to irradiation with particle-emitting radioligands. Spheroid growth was monitored over 35 days. N = 38090 images, acquired within seven experiments and two studies, were included. Spheroids were delineated with a threshold-based method followed by manual corrections and the resulting delineations served as ground truth for network training and testing. The data were divided into two independent data sets: one for training and internal validation using a 5-fold cross-validation (N = 21567; main data set) and another for final independent testing (N = 16523). The network was developed using the nnU-Net v2 deep-learning (DL) framework. DL-based and manual delineations were compared using the Dice similarity coefficient (DSC). Additionally, treatment effects in a spheroid experiment were compared by quantifying half-maximum spheroid control doses (SCD50). The median DSC values in the main and test data sets were 0.979 and 0.974, respectively. In the main data set, only 7% (N = 1571) of the DL-generated delineations and 8% (N = 1304) in the test data set showed DSC < 0.9, indicating high performance. The SCD50 values were comparable between manual (day 13: 0.086 ± 0.001, day 35: 0.150 ± 0.001) and DL-based delineations (day 13: 0.083 ± 0.002, day 35: 0.149 ± 0.007). The network enables fast and accurate delineation of tumor spheroids in treatment response assays, reducing the time needed to delineate all spheroid images of a single experiment from several days with the previously applied method to a few hours only.

The [18F]FDG-PET-derived total metabolic tumor volume (TMTV) has a high prognostic value in patients with Hodgkin and Non-Hodgkin lymphoma. However, in order to enable TMTV as a biomarker for clinical use, an accurate and fast method of tumor delineation in lymphoma patients is needed. Deep-learning-based methods have shown promising results in this field and offer distinct advantages over classical approaches. Therefore, the goal of this work was to train a convolutional neural network (CNN) for delineation of all lymphoma lesions regardless of their size and uptake characteristics while performing the optimal contouring of each individual lesion. A neural network was trained with the nnU-Net software package. A total of 1192 [18F]FDG-PET/CT scans from 716 patients with Non-Hodgkin lymphoma participating in the PETAL trial comprised the main dataset which was used for training. The ground truth delineation included all lesions that were clinically considered as lymphoma manifestations by an experienced observer and was developed iteratively with the assistance of intermediate CNN models. Performance of the trained network was assessed in the main dataset via 5-fold cross-validation as well as in external benchmark dataset (N = 60 scans). Comparing the manual and automated delineations in the main (external) dataset, the aggregated Dice coefficient reached 0.895 (0.715) and the corresponding TMTVs were highly correlated with R2 = 0.974 (0.767). The main (external) dataset contained a total of 8971 (713) manually delineated lesions, the detection sensitivity of which was 71.2

A reduction of dose and/or acquisition duration of PET examinations is desirable in terms of radiation protection, patient comfort and throughput, but leads to decreased image quality due to poorer image statistics. Recently, different deep-learning based methods have been proposed to improve image quality of low-count PET images. For example, one such approach allows the generation of AI-enhanced PET images (AI-PET) based on ultra-low count PET/CT scans. The performance of this algorithm has so far only been clinically evaluated on patient data featuring limited scan statistics and unknown actual activity concentration. Therefore, this study investigates the performance of this deep-learning algorithm using PET measurements of a phantom resembling different lesion sizes and count statistics (from ultra-low to high) to understand the capabilities and limitations of AI-based post processing for improved image quality in ultra-low count PET imaging. A previously trained pix2pixHD Generative Adversarial Network was evaluated. To this end, a NEMA PET body phantom filled with two sphere-to-background activity concentration ratios (4:1 and 10:1) and two attenuation scenarios to investigate the effects of obese patients was scanned in list mode. Images were reconstructed with 13 different acquisition durations ranging from 5 s up to 900 s. Image noise, recovery coefficients, SUV-differences, image quality measurement metrics such as the Structural Similarity Index Metric, and the contrast-to-noise-ratio were assessed. In addition, the benefits of the deep-learning network over Gaussian smoothing were investigated. The presented AI-algorithm is very well suitable for denoising ultra-low count PET images and for restoring structural information, but increases image noise in ultra-high count PET scans. The generated AI-PET scans strongly underestimate SUV especially in small lesions with a diameter ≤ 17 mm, while quantitative measures of large lesions ≥ 37 mm in diameter were accurately recovered. In ultra-low count or low contrast images, the AI algorithm might not be able to recognize small lesions ≤ 13 mm in diameter. In comparison to standardized image post-processing using a Gaussian filter, the deep-learning network is better suited to improve image quality, but at the same time degrades SUV accuracy to a greater extent than post-filtering and quantitative SUV accuracy varies for different lesion sizes. Phantom-based validation of AI-based algorithms allows for a detailed assessment of the performance, limitations, and generalizability of deep-learning based algorithms for PET image enhancement. Here it was confirmed that the AI-based approach performs very well in denoising ultra-low count PET images and outperforms traditional Gaussian post-filtering. However, there are strong limitations in terms of quantitative accuracy and detectability of small lesions.

Glioblastoma (GBM) is the most aggressive form of malignant gliomas, with the eicosanoid-synthesizing enzyme cyclooxygenase-2 (COX-2) playing a pivotal role in its progression via the COX-2/prostaglandin E2/4 axis. COX-2 upregulations in tumor cells induces a pro-inflammatory tumor microenvironment (TME), affecting the behavior of invading bone marrow-derived macrophages (Mϕ) and brain-resident microglia (MG) through unclear autocrine and paracrine mechanisms. Using CRISPR/Cas9 technology, we generated COX-2 knockout U87 glioblastoma cells. In spheroids and in vivo xenografts, this resulted in a significant inhibition of tumorigenic properties, while not observed in standard adherent monolayer culture. Here, the knockout induced a G1 cell cycle arrest in adherent cells, accompanied by increased ROS, mitochondrial activity, and cytochrome c-mediated apoptosis. In spheroids and xenograft models, COX-2 knockout led to notable growth delays and increased cell death, characterized by features of both apoptosis and autophagy. Interestingly, these effects were partially reversed in subcutaneous xenografts after co-culture with Mϕ, while co-culture with MG enhanced the growth-suppressive effects. In an orthotopic model, COX-2 knockout tumors displayed reduced proliferation (fewer Ki-67 positive cells), increased numbers of GFAP-positive astrocytes, and signs of membrane blebbing. These findings highlight the potential of COX-2 knockout and suppression as a therapeutic strategy in GBM, particularly when combined with suppression of infiltrating macrophages and stabilization of resident microglia populations to enhance anti-tumor effects.

The inducible enzyme cyclooxygenase-2 (COX-2) and the subsequent synthesis of eicosanoids initiated by this enzyme are important molecular players in bone healing. In this pilot study, the suitability of a novel selective COX-2 inhibitor bearing a nitric oxide (NO)-releasing moiety was investigated as a modulator of healing a critical-size bone defect in rats. A 5 mm femoral defect was randomly filled with no material (negative control, NC), a mixture of collagen and autologous bone fragments (positive control, PC), or polycaprolactone-co-lactide (PCL)-scaffolds coated with two types of artificial extracellular matrix (aECM; collagen/chondroitin sulfate (Col/CS) or collagen/polysulfated hyaluronic acid (Col/sHA3)). Bone healing was monitored by a dual-tracer ([18F]FDG/[18F]fluoride) approach using PET/CT imaging in vivo. In addition, ex vivo µCT imaging as well as histological and immunohistochemical studies were performed 16 weeks post-surgery. A significant higher uptake of [18F]FDG, a surrogate marker for inflammatory infiltrate, but not of [18F]fluoride, representing bone mineralization, was observed in the implanted PCL-scaffolds coated with either Col/CS or Col/sHA3. Molecular targeting of COX-2 with NO-coxib had no significant effect on tracer uptake in any of the groups. Histological and immunohistochemical staining showed no evidence of a positive or negative influence of NO-coxib treatment on bone healing.

Tumor asphericity in 18F-FDG PET is a prognostic marker that has been investigated in small pilot studies of patients with head and neck squamous cell carcinoma (HNSCC). Here, we investigated the prognostic role of asphericity in a large multicenter database of patients with HNSCC treated with primary radiotherapy or chemoradiation and assessed its independent prognostic value. Methods: In total, 1,104 patients were included in this analysis. All received pretreatment 18F-FDG PET scans. Clinical risk factors were evaluated, and quantitative PET parameters SUVmax, metabolic tumor volume (MTV), total lesion glycolysis, and asphericity were calculated. Primary study endpoints were overall survival (OS) and locoregional control (LRC). Uni- and multivariate Cox regression analyses were performed. Additionally, asphericity was combined with the best-established quantitative PET parameter of MTV, and the combinatory approach of using asphericity and MTV was compared with the use of only asphericity or MTV by bootstrap analyses. Results: Asphericity showed only a modest correlation with the established PET parameters of MTV, SUVmax, and total lesion glycolysis. On univariate testing asphericity was strongly associated with the outcome of patients (LRC and OS with P < 0.001). In multivariate testing of all imaging parameters that were not highly correlated, both MTV and asphericity showed a significant association with LRC (P < 0.001 for MTV and P = 0.021 for asphericity) and OS (P < 0.001 for MTV and asphericity). Asphericity and MTV were binarized and combined for risk stratification, and the prognostic value of the combination was compared with the prognostic value of individual parameters. Bootstrapping revealed significantly better performance by the combinatory approach when compared with MTV (P = 0.012 for LRC and P < 0.001 for OS) and asphericity with regard to OS (P < 0.001) but not for LRC (P = 0.53). Conclusion: We were able to show that asphericity bears independent prognostic value and significantly improves risk stratification when combined with MTV in a comprehensive retrospective cohort of HNSCC patients.

BackgroundOropharyngeal carcinomas are characterized by an increasing incidence and a relatively good prognosis. Nonetheless, a considerable number of patients develops metachronous distant metastases; identification of these patients is an urgent medical need. MethodsThis is a retrospective multicenter evaluation of 431 patients. All patients underwent [18F]-FDG positron emission tomography (PET). The cohort was split into an explorative group (n = 366) and a validation group (n = 65). Lesions were manually delineated in the explorative group and automatically delineated by a convolutional neuronal network (CNN) in the validation group. Quantitative PET parameters standardized uptake value (SUV), metabolic tumor volume (MTV), and total lesion glycolysis (TLG) were calculated for primary tumors (prim) and tumor plus lymphnodes (all). Association of parameters with freedom from distant metastases (FFDM) and overall survival (OS) was tested by cox regression analyses. ResultsIn the explorative group, univariate analyses revealed an association of metric MTVprim (p = 0.022), MTVall (p < 0.001) and TLGall (p < 0.001) with FFDM, binarized parameters were also associated with FFDM (p < 0.001 and p = 0.002). Bootstrap analyses revealed a significantly better association of TLGall compared to TLGprim with FFDM (p = 0.02). MTVall and TLGall remained significantly associated with FFDM upon multivariate testing (p = 0.002, p = 0.031, respectively). In the validation group, the cutoff value for TLGall but not for TLGprim was significantly associated with FFDM (HR = 3.1, p = 0.045). Additional analyses with manually delineated contours of the validation cohort revealed a similar effect (HR = 3.47, p = 0.026). No considerable differences between HPV positive and negative disease were observed. ConclusionsTLGall is a promising biomarker to select OPC patients with high risk for metachronous distant metastases.

Abstract Background PSMA-PET is increasingly used for staging prostate cancer (PCA) patients. However, it is not clear if quantitative imaging parameters of positron emission tomography (PET) have an impact on disease progression and are thus important for the prognosis of localized PCA. Methods This is a monocenter retrospective analysis of 86 consecutive patients with localized intermediate or high-risk PCA and PSMA-PET before treatment The quantitative PET parameters maximum standardized uptake value (SUVmax), tumor asphericity (ASP), PSMA tumor volume (PSMA-TV), and PSMA total lesion uptake (PSMA-TLU = PSMA-TV × SUVmean) were assessed for their prognostic significance in patients with radiotherapy or surgery. Cox regression analyses were performed for biochemical recurrence-free survival, overall survival (OS), local control, and loco-regional control (LRC). Results 67% of patients had high-risk disease, 51 patients were treated with radiotherapy, and 35 with surgery. Analysis of metric PET parameters in the whole cohort revealed a significant association of PSMA-TV (p = 0.003), PSMA-TLU (p = 0.004), and ASP (p < 0.001) with OS. Upon binarization of PET parameters, several other parameters showed a significant association with clinical outcome. When analyzing high-risk patients according to the primary treatment approach, a previously published cut-off for SUVmax (8.6) showed a significant association with LRC in surgically treated (p = 0.048), but not in primary irradiated (p = 0.34) patients. In addition, PSMA-TLU (p = 0.016) seemed to be a very promising biomarker to stratify surgical patients. Conclusion Our data confirm one previous publication on the prognostic impact of SUVmax in surgically treated patients with high-risk PCA. Our exploratory analysis indicates that PSMA-TLU might be even better suited. The missing association with primary irradiated patients needs prospective validation with a larger sample size to conclude a predictive potential. Trial registration Due to the retrospective nature of this research, no registration was carried out.

Abstract Background Residual image noise is substantial in positron emission tomography (PET) and one of the factors limiting lesion detection, quantification, and overall image quality. Thus, improving noise reduction remains of considerable interest. This is especially true for respiratory-gated PET investigations. The only broadly used approach for noise reduction in PET imaging has been the application of low-pass filters, usually Gaussians, which however leads to loss of spatial resolution and increased partial volume effects affecting detectability of small lesions and quantitative data evaluation. The bilateral filter (BF) — a locally adaptive image filter — allows to reduce image noise while preserving well defined object edges but manual optimization of the filter parameters for a given PET scan can be tedious and time-consuming, hampering its clinical use. In this work we have investigated to what extent a suitable deep learning based approach can resolve this issue by training a suitable network with the target of reproducing the results of manually adjusted case-specific bilateral filtering. Methods Altogether, 69 respiratory-gated clinical PET/CT scans with three different tracers ( $$[^{18}\text {F}]$$ [ 18 F ] FDG, $$[^{18}\text {F}]$$ [ 18 F ] L-DOPA, $$[^{68}\text {Ga}]$$ [ 68 Ga ] DOTATATE) were used for the present investigation. Prior to data processing, the gated data sets were split, resulting in a total of 552 single-gate image volumes. For each of these image volumes, four 3D ROIs were delineated: one ROI for image noise assessment and three ROIs for focal uptake (e.g. tumor lesions) measurements at different target/background contrast levels. An automated procedure was used to perform a brute force search of the two-dimensional BF parameter space for each data set to identify the “optimal” filter parameters to generate user-approved ground truth input data consisting of pairs of original and optimally BF filtered images. For reproducing the optimal BF filtering, we employed a modified 3D U-Net CNN incorporating residual learning principle. The network training and evaluation was performed using a 5-fold cross-validation scheme. The influence of filtering on lesion SUV quantification and image noise level was assessed by calculating absolute and fractional differences between the CNN, manual BF, or original (STD) data sets in the previously defined ROIs. Results The automated procedure used for filter parameter determination chose adequate filter parameters for the majority of the data sets with only 19 patient data sets requiring manual tuning. Evaluation of the focal uptake ROIs revealed that CNN as well as BF based filtering essentially maintain the focal $$\text {SUV}_\text {max}$$ SUV max values of the unfiltered images with a low mean ± SD difference of $$\delta \text {SUV}_\text {max}^{\text {CNN},\text {STD}}$$ δ SUV max CNN , STD = (−3.9 ± 5.2)% and $$\delta \text {SUV}_\text {max}^{\text {BF},\text {STD}}$$ δ SUV max BF , STD = (−4.4 ± 5.3)%. Regarding relative performance of CNN versus BF, both methods lead to very similar $$\text {SUV}_\text {max}$$ SUV max values in the vast majority of cases with an overall average difference of $$\delta \text {SUV}_\text {max}^{\text {CNN},\text {BF}}$$ δ SUV max CNN , BF = (0.5 ± 4.8)%. Evaluation of the noise properties showed that CNN filtering mostly satisfactorily reproduces the noise level and characteristics of BF with $$\delta \text {Noise}^{\text {CNN},\text {BF}}$$ δ Noise CNN , BF = (5.6 ± 10.5)%. No significant tracer dependent differences between CNN and BF were observed. Conclusions Our results show that a neural network based denoising can reproduce the results of a case by case optimized BF in a fully automated way. Apart from rare cases it led to images of practically identical quality regarding noise level, edge preservation, and signal recovery. We believe such a network might proof especially useful in the context of improved motion correction of respiratory-gated PET studies but could also help to establish BF-equivalent edge-preserving CNN filtering in clinical PET since it obviates time consuming manual BF parameter tuning.

Purpose: Ga-68-labeled fibroblast activation protein inhibitor (FAPI) is a novel PET tracer with great potential for staging pancreatic cancer. Data on locally advanced or recurrent disease is sparse, especially on tracer uptake before and after high dose chemoradiotherapy (CRT). The aim of this study was to evaluate [Ga-68]Ga-FAPI-46 PET/CT staging in this setting. Methods: Twenty-seven patients with locally recurrent or locally advanced pancreatic adenocarcinoma (LRPAC n = 15, LAPAC n = 12) in stable disease or partial remission after chemotherapy underwent FAPI PET/CT and received consolidation CRT in stage M0 with follow-up FAPI PET/CT every three months until systemic progression. Quantitative PET parameters SUVmax, SUVmean, FAPI-derived tumor volume and total lesion FAPI-uptake were measured in baseline and follow-up PET/CT scans. Contrast-enhanced CT (ceCT) and PET/CT data were evaluated blinded and staged according to TNM classification. Results: FAPI PET/CT modified staging compared to ceCT alone in 23 of 27 patients in baseline, resulting in major treatment alterations in 52% of all patients (30%: target volume adjustment due to N downstaging, 15%: switch to palliative systemic chemotherapy only due to diffuse metastases, 7%: abortion of radiotherapy due to other reasons). Regarding follow-up scans, major treatment alterations after performing FAPI PET/CT were noted in eleven of 24 follow-up scans (46%) with switch to systemic chemotherapy or best supportive care due to M upstaging and ablative radiotherapy of distant lymph node and oligometastasis. Unexpectedly, in more than 90 % of the follow-up scans, radiotherapy did not induce local fibrosis related FAPI uptake. During the first follow-up, all quantitative PET metrics decreased, and irradiated lesions showed significantly lower FAPI uptake in locally controlled disease (SUVmax p = 0.047, SUVmean p = 0.0092) compared to local failure. Conclusion: Compared to ceCT, FAPI PET/CT led to major therapeutic alterations in patients with LRPAC and LAPAC prior to and after radiotherapy, which might help identify patients benefiting from adjustments in every treatment stage. FAPI PET/CT should be considered a useful diagnostic tool in LRPAC or LAPAC before and after CRT.

Background: The aim of this retrospective study was to assess the value of F-18-fluoro-2-deoxy-D-glucose positron emission tomography/computed tomography ([F-18]FDG PET/CT parameters in cN1-cN3 non-small cell lung cancer (NSCLC) patients. Materials and methods: 59 consecutive patients (35 M, 24 F) with NSCLC who underwent pretreatment [F-18]FDG PET/CT were enrolled to this study. Several primary tumor PET parameters, including the maximum and mean standardized uptake value (SUVmax and SUVmean) , the metabolic active tumor volume (MTV) and the total lesion glycolysis (TLG = MTVxSUV(mean)), were extracted and analysed. Overall survival was defined as time from primary diagnosis to death or the last info. Results: In the whole analysed group 44 patients underwent curative treatment, while 15, because of the severity of the disease, were classified for palliative treatment. Univariate Cox analysis of clinical and metric PET parameters revealed that MTV was a significant prognostic factor for OS (p = 0.024), while TLG and curative treatment showed a trend for significance (p < 0.1). In multivariate Cox regression (MTV and curative treatment) MTV remained a significant factor (p = 0.047). Conclusions: Metabolic tumor volume of the primary tumor was the only independent prognostic factor for cN1-cN3 NSCLC patients.

The objective of this study was to assess the prognostic value of asphericity (ASP) and standardized uptake ratio (SUR) in cervical cancer patients. Retrospective analysis was performed on a group of 508 (aged 55 ± 12 years) previously untreated cervical cancer patients. All patients underwent a pretreatment [ 18 F]FDG PET/CT study to assess the severity of the disease. The metabolic tumor volume (MTV) of the cervical cancer was delineated with an adaptive threshold method. For the resulting ROIs the maximum standardized uptake value (SUV max ) was measured. In addition, ASP and SUR were determined as previously described. Univariate Cox regression and Kaplan–Meier analysis with respect to event free survival (EFS), overall survival (OS), freedom from distant metastasis (FFDM) and locoregional control (LRC) was performed. Additionally, a multivariate Cox regression including clinically relevant parameters was performed. In the survival analysis, MTV and ASP were shown to be prognostic factors for all investigated endpoints. Tumor metabolism quantified with the SUV max was not prognostic for any of the endpoints (p > 0.2). The SUR did not reach statistical significance either (p = 0.1, 0.25, 0.066, 0.053, respectively). In the multivariate analysis, the ASP remained a significant factor for EFS and LRC, while MTV was a significant factor for FFDM, indicating their independent prognostic value for the respective endpoints. The alternative parameter ASP has the potential to improve the prognostic value of [ 18 F]FDG PET/CT for event-free survival and locoregional control in radically treated cervical cancer patients.

Ziel/Aim Breathing related patient motion is a serious source of image blurring in oncological whole body PET. Respiratory gating allows to subdivide the aquisition data into temporal bins ("gates") depending on the breathing cycle which are basically motion-free but exhhibit increased noise levels. Image registration of all gates to a reference gate and subsequent averaging is possible but traditional registration algorithms are frequently slow or of limited accuracy.

Background Abnormal activity of 18 F-FDG PET/CT is a major Duke criterion in the diagnostic work-up of infective prosthetic valve endocarditis (IE). We hypothesized that quantitative lesion assessment by 18 F-FDG PET/CT-derived standard maximum uptake ratio (SURmax), metabolic volume (MV), and total lesion glycolysis (TLG) might be useful in distinct subgroups of IE patients (e.g. IE-related abscess formation). Methods All patients ( n = 27) hospitalized in our tertiary IE referral medical center from January 2014 to October 2018 with preoperatively performed 18 F-FDG PET/CT and surgically confirmed IE were included into this retrospective analysis. Results Patients with surgically confirmed abscess formation ( n = 10) had significantly increased MV (by ~ fivefold) and TLG (by ~ sevenfold) as compared to patients without abscess ( n = 17). Receiver operation characteristics (ROC) analyses demonstrated that TLG (calculated as MV × SURmean, i.e. TLG (SUR)) had the most favorable area under the ROC curve (0.841 [CI 0.659 to 1.000]) in predicting IE-related abscess formation. This resulted in a sensitivity of 80% and a specificity of 88% at a cut-off value of 14.14 mL for TLG (SUR). Conclusion We suggest that 18 F-FDG PET/CT-derived quantitative assessment of TLG (SUR) may provide a novel diagnostic tool in predicting endocarditis-associated abscess formation.