Assessment of [18F]PI-2620 Tau-PET Quantification via Non-Invasive Automatized Image Derived Input Function.

PURPOSE: [18F]PI-2620 positron emission tomography (PET) detects misfolded tau in progressive supranuclear palsy (PSP) and Alzheimer's disease (AD). We questioned the feasibility and value of absolute [18F]PI-2620 PET quantification for assessing tau by regional distribution volumes (VT). Here, arterial input functions (AIF) represent the gold standard, but cannot be applied in routine clinical practice, whereas image-derived input functions (IDIF) represent a non-invasive alternative. We aimed to validate IDIF against AIF and we evaluated the potential to discriminate patients with PSP and AD from healthy controls by non-invasive quantification of [18F] PET. METHODS: In the first part of the study, we validated AIF derived from radial artery whole blood against IDIF by investigating 20 subjects (ten controls and ten patients). IDIF were generated by manual extraction of the carotid artery using the average and the five highest (max5) voxel intensity values and by automated extraction of the carotid artery using the average and the maximum voxel intensity value. In the second part of the study, IDIF quantification using the IDIF with the closest match to the AIF was transferred to group comparison of a large independent cohort of 40 subjects (15 healthy controls, 15 PSP patients and 10 AD patients). We compared VT and VT ratios, both calculated by Logan plots, with distribution volume (DV) ratios using simplified reference tissue modelling and standardized uptake value (SUV) ratios. RESULTS: AIF and IDIF showed highly correlated input curves for all applied IDIF extraction methods (0.78 < r < 0.83, all p < 0.0001; area under the curves (AUC): 0.73 < r ≤ 0.82, all p ≤ 0.0003). Regarding the VT values, correlations were mainly found between those generated by the AIF and by the IDIF methods using the maximum voxel intensity values. Lowest relative differences (RD) were observed by applying the manual method using the five highest voxel intensity values (max5) (AIF vs. IDIF manual, avg: RD = -82%; AIF vs. IDIF automated, avg: RD = -86%; AIF vs. IDIF manual, max5: RD = -6%; AIF vs. IDIF automated, max: RD = -26%). Regional VT values revealed considerable variance at group level, which was strongly reduced upon scaling by the inferior cerebellum. The resulting VT ratio values were adequate to detect group differences between patients with PSP or AD and healthy controls (HC) (PSP target region (globus pallidus): HC vs. PSP vs. AD: 1.18 vs. 1.32 vs. 1.16; AD target region (Braak region I): HC vs. PSP vs. AD: 1.00 vs. 1.00 vs. 1.22). VT ratios and DV ratios outperformed SUV ratios and VT in detecting differences between PSP and healthy controls, whereas all quantification approaches performed similarly in comparing AD and healthy controls. CONCLUSION: Blood-free IDIF is a promising approach for quantification of [18F]PI-2620 PET, serving as correlating surrogate for invasive continuous arterial blood sampling. Regional [18F]PI-2620 VT show large variance, in contrast to regional [18F]PI-2620 VT ratios scaled with the inferior cerebellum, which are appropriate for discriminating PSP, AD and healthy controls. DV ratios obtained by simplified reference tissue modeling are similarly suitable for this purpose.

Validity and value of metabolic connectivity in mouse models of ß-amyloid and tauopathy.

Among functional imaging methods, metabolic connectivity (MC) is increasingly used for investigation of regional network changes to examine the pathophysiology of neurodegenerative diseases such as Alzheimer's disease (AD) or movement disorders. Hitherto, MC was mostly used in clinical studies, but only a few studies demonstrated the usefulness of MC in the rodent brain. The goal of the current work was to analyze and validate metabolic regional network alterations in three different mouse models of neurodegenerative diseases (ß-amyloid and tau) by use of 2-deoxy-2-[18F]fluoro-d-glucose positron emission tomography (FDG-PET) imaging. We compared the results of FDG-µPET MC with conventional VOI-based analysis and behavioral assessment in the Morris water maze (MWM). The impact of awake versus anesthesia conditions on MC read-outs was studied and the robustness of MC data deriving from different scanners was tested. MC proved to be an accurate and robust indicator of functional connectivity loss when sample sizes ≥12 were considered. MC readouts were robust across scanners and in awake/ anesthesia conditions. MC loss was observed throughout all brain regions in tauopathy mice, whereas ß-amyloid indicated MC loss mainly in spatial learning areas and subcortical networks. This study established a methodological basis for the utilization of MC in different ß-amyloid and tau mouse models. MC has the potential to serve as a read-out of pathological changes within neuronal networks in these models.

Deciphering sources of PET signals in the tumor microenvironment of glioblastoma at cellular resolution.

Various cellular sources hamper interpretation of positron emission tomography (PET) biomarkers in the tumor microenvironment (TME). We developed an approach of immunomagnetic cell sorting after in vivo radiotracer injection (scRadiotracing) with three-dimensional (3D) histology to dissect the cellular allocation of PET signals in the TME. In mice with implanted glioblastoma, translocator protein (TSPO) radiotracer uptake per tumor cell was higher compared to tumor-associated microglia/macrophages (TAMs), validated by protein levels. Translation of in vitro scRadiotracing to patients with glioma immediately after tumor resection confirmed higher single-cell TSPO tracer uptake of tumor cells compared to immune cells. Across species, cellular radiotracer uptake explained the heterogeneity of individual TSPO-PET signals. In consideration of cellular tracer uptake and cell type abundance, tumor cells were the main contributor to TSPO enrichment in glioblastoma; however, proteomics identified potential PET targets highly specific for TAMs. Combining cellular tracer uptake measures with 3D histology facilitates precise allocation of PET signals and serves to validate emerging novel TAM-specific radioligands.

Distinct molecular profiles of skull bone marrow in health and neurological disorders.

The bone marrow in the skull is important for shaping immune responses in the brain and meninges, but its molecular makeup among bones and relevance in human diseases remain unclear. Here, we show that the mouse skull has the most distinct transcriptomic profile compared with other bones in states of health and injury, characterized by a late-stage neutrophil phenotype. In humans, proteome analysis reveals that the skull marrow is the most distinct, with differentially expressed neutrophil-related pathways and a unique synaptic protein signature. 3D imaging demonstrates the structural and cellular details of human skull-meninges connections (SMCs) compared with veins. Last, using translocator protein positron emission tomography (TSPO-PET) imaging, we show that the skull bone marrow reflects inflammatory brain responses with a disease-specific spatial distribution in patients with various neurological disorders. The unique molecular profile and anatomical and functional connections of the skull show its potential as a site for diagnosing, monitoring, and treating brain diseases.

[18F]F-DED PET imaging of reactive astrogliosis in neurodegenerative diseases: preclinical proof of concept and first-in-human data.

OBJECTIVES: Reactive gliosis is a common pathological hallmark of CNS pathology resulting from neurodegeneration and neuroinflammation. In this study we investigate the capability of a novel monoamine oxidase B (MAO-B) PET ligand to monitor reactive astrogliosis in a transgenic mouse model of Alzheimer`s disease (AD). Furthermore, we performed a pilot study in patients with a range of neurodegenerative and neuroinflammatory conditions. METHODS: A cross-sectional cohort of 24 transgenic (PS2APP) and 25 wild-type mice (age range: 4.3-21.0 months) underwent 60 min dynamic [18F]fluorodeprenyl-D2 ([18F]F-DED), static 18 kDa translocator protein (TSPO, [18F]GE-180) and ß-amyloid ([18F]florbetaben) PET imaging. Quantification was performed via image derived input function (IDIF, cardiac input), simplified non-invasive reference tissue modelling (SRTM2, DVR) and late-phase standardized uptake value ratios (SUVr). Immunohistochemical (IHC) analyses of glial fibrillary acidic protein (GFAP) and MAO-B were performed to validate PET imaging by gold standard assessments. Patients belonging to the Alzheimer's disease continuum (AD, n = 2), Parkinson's disease (PD, n = 2), multiple system atrophy (MSA, n = 2), autoimmune encephalitis (n = 1), oligodendroglioma (n = 1) and one healthy control underwent 60 min dynamic [18F]F-DED PET and the data were analyzed using equivalent quantification strategies. RESULTS: We selected the cerebellum as a pseudo-reference region based on the immunohistochemical comparison of age-matched PS2APP and WT mice. Subsequent PET imaging revealed that PS2APP mice showed elevated hippocampal and thalamic [18F]F-DED DVR when compared to age-matched WT mice at 5 months (thalamus: + 4.3%; p = 0.048), 13 months (hippocampus: + 7.6%, p = 0.022) and 19 months (hippocampus: + 12.3%, p < 0.0001; thalamus: + 15.2%, p < 0.0001). Specific [18F]F-DED DVR increases of PS2APP mice occurred earlier when compared to signal alterations in TSPO and ß-amyloid PET and [18F]F-DED DVR correlated with quantitative immunohistochemistry (hippocampus: R = 0.720, p < 0.001; thalamus: R = 0.727, p = 0.002). Preliminary experience in patients showed [18F]F-DED VT and SUVr patterns, matching the expected topology of reactive astrogliosis in neurodegenerative (MSA) and neuroinflammatory conditions, whereas the patient with oligodendroglioma and the healthy control indicated [18F]F-DED binding following the known physiological MAO-B expression in brain. CONCLUSIONS: [18F]F-DED PET imaging is a promising approach to assess reactive astrogliosis in AD mouse models and patients with neurological diseases.

Depletion and activation of microglia impact metabolic connectivity of the mouse brain.

AIM: We aimed to investigate the impact of microglial activity and microglial FDG uptake on metabolic connectivity, since microglial activation states determine FDG-PET alterations. Metabolic connectivity refers to a concept of interacting metabolic brain regions and receives growing interest in approaching complex cerebral metabolic networks in neurodegenerative diseases. However, underlying sources of metabolic connectivity remain to be elucidated. MATERIALS AND METHODS: We analyzed metabolic networks measured by interregional correlation coefficients (ICCs) of FDG-PET scans in WT mice and in mice with mutations in progranulin (Grn) or triggering receptor expressed on myeloid cells 2 (Trem2) knockouts (-/-) as well as in double mutant Grn-/-/Trem2-/- mice. We selected those rodent models as they represent opposite microglial signatures with disease associated microglia in Grn-/- mice and microglia locked in a homeostatic state in Trem2-/- mice; however, both resulting in lower glucose uptake of the brain. The direct influence of microglia on metabolic networks was further determined by microglia depletion using a CSF1R inhibitor in WT mice at two different ages. Within maps of global mean scaled regional FDG uptake, 24 pre-established volumes of interest were applied and assigned to either cortical or subcortical networks. ICCs of all region pairs were calculated and z-transformed prior to group comparisons. FDG uptake of neurons, microglia, and astrocytes was determined in Grn-/- and WT mice via assessment of single cell tracer uptake (scRadiotracing). RESULTS: Microglia depletion by CSF1R inhibition resulted in a strong decrease of metabolic connectivity defined by decrease of mean cortical ICCs in WT mice at both ages studied (6-7 m; p = 0.0148, 9-10 m; p = 0.0191), when compared to vehicle-treated age-matched WT mice. Grn-/-, Trem2-/- and Grn-/-/Trem2-/- mice all displayed reduced FDG-PET signals when compared to WT mice. However, when analyzing metabolic networks, a distinct increase of ICCs was observed in Grn-/- mice when compared to WT mice in cortical (p < 0.0001) and hippocampal (p < 0.0001) networks. In contrast, Trem2-/- mice did not show significant alterations in metabolic connectivity when compared to WT. Furthermore, the increased metabolic connectivity in Grn-/- mice was completely suppressed in Grn-/-/Trem2-/- mice. Grn-/- mice exhibited a severe loss of neuronal FDG uptake (- 61%, p < 0.0001) which shifted allocation of cellular brain FDG uptake to microglia (42% in Grn-/- vs. 22% in WT). CONCLUSIONS: Presence, absence, and activation of microglia have a strong impact on metabolic connectivity of the mouse brain. Enhanced metabolic connectivity is associated with increased microglial FDG allocation.

Machine learning-based approach reveals essential features for simplified TSPO PET quantification in ischemic stroke patients.

INTRODUCTION: Neuroinflammation evaluation after acute ischemic stroke is a promising option for selecting an appropriate post-stroke treatment strategy. To assess neuroinflammation in vivo, translocator protein PET (TSPO PET) can be used. However, the gold standard TSPO PET quantification method includes a 90 min scan and continuous arterial blood sampling, which is challenging to perform on a routine basis. In this work, we determine what information is required for a simplified quantification approach using a machine learning algorithm. MATERIALS AND METHODS: We analyzed data from 18 patients with ischemic stroke who received 0-90 min [18F]GE-180 PET as well as T1-weigted (T1w), FLAIR, and arterial spin labeling (ASL) MRI scans. During PET scans, five manual venous blood samples at 5, 15, 30, 60, and 85 min post injection (p.i.) were drawn, and plasma activity concentration was measured. Total distribution volume (VT) was calculated using Logan plot with the full dynamic PET and an image-derived input function (IDIF) from the carotid arteries. IDIF was scaled by a calibration factor derived from all the measured plasma activity concentrations. The calculated VT values were used for training a random forest regressor. As input features for the model, we used three late PET frames (60-70, 70-80, and 80-90 min p.i.), the ASL image reflecting perfusion, the voxel coordinates, the lesion mask, and the five plasma activity concentrations. The algorithm was validated with the leave-one-out approach. To estimate the impact of the individual features on the algorithm's performance, we used Shapley Additive Explanations (SHAP). Having determined that the three late PET frames and the plasma activity concentrations were the most important features, we tested a simplified quantification approach consisting of dividing a late PET frame by a plasma activity concentration. All the combinations of frames/samples were compared by means of concordance correlation coefficient and Bland-Altman plots. RESULTS: When using all the input features, the algorithm predicted VT values with high accuracy (87.8 ±â€¯8.3%) for both lesion and non-lesion voxels. The SHAP values demonstrated high impact of the late PET frames (60-70, 70-80, and 80-90 min p.i.) and plasma activity concentrations on the VT prediction, while the influence of the ASL-derived perfusion, voxel coordinates, and the lesion mask was low. Among all the combinations of the late PET frames and plasma activity concentrations, the 70-80 min p.i. frame divided by the 30 min p.i. plasma sample produced the closest VT estimate in the ischemic lesion. CONCLUSION: Reliable TSPO PET quantification is achievable by using a single late PET frame divided by a late blood sample activity concentration.

Detectors in positron emission tomography.

Positron emission tomography is a highly sensitive molecular imaging modality, based on the coincident detection of annihilation photons after positron decay. The most used detector is based on dense, fast, and luminous scintillators read out by light sensors. This review covers the various detector concepts for clinical and preclinical systems.

18 kDa translocator protein positron emission tomography facilitates early and robust tumor detection in the immunocompetent SB28 glioblastoma mouse model.

Introduction: The 18 kDa translocator protein (TSPO) receives growing interest as a biomarker in glioblastoma. Mouse models can serve as an important tool for the investigation of biomarkers in glioblastoma, but several glioblastoma models indicated only low TSPO-PET signals in contrast to high TSPO-PET signals of human glioblastoma. Thus, we aimed to investigate TSPO-PET imaging in the syngeneic immunocompetent SB28 mouse model, which is thought to closely represent the tumor microenvironment (TME) of human glioblastoma. Methods: Dynamic TSPO-PET/CT imaging was performed for 60 min after injection of 13.6 ± 4.2 MBq [18F]GE-180. Contrast enhanced CT (ceCT) was acquired prior to PET and served for assessment of tumor volumes and attenuation correction. SB28 and sham mice were imaged at an early (week-1; n = 6 SB28, n = 6 sham) and a late time-point (week-3; n = 8 SB28, n = 9 sham) after inoculation. Standard of truth ex vivo tumor volumes were obtained for SB28 mice at the late time-point. Tracer kinetics were analyzed for the lesion site and the carotid arteries to establish an image derived input function (IDIF). TSPO-PET and ceCT lesion volumes were compared with ex vivo volumes by calculation of root-mean-square-errors (RMSE). Volumes of distribution (VTmax/mean) in the lesion were calculated using carotid IDIF and standardized uptake values (SUVmax/mean) were obtained for a 40-60 min time frame. Results: Higher uptake rate constants (K1) were observed for week-1 SB28 tumor lesions when compared to week-3 SB28 tumor lesions. Highest agreement between TSPO-PET lesion volumes and ex vivo tumor volumes was achieved with a 50% maximum threshold (RMSE-VT: 39.7%; RMSE-SUV: 34.4%), similar to the agreement of ceCT tumor volumes (RMSE: 30.1%). Lesions of SB28 mice had higher PET signal when compared to sham mice at week-1 (VTmax 6.6 ± 2.9 vs. 3.9 ± 0.8, p = 0.035; SUVmax 2.3 ± 0.5 vs. 1.2 ± 0.1, p < 0.001) and PET signals remained at a similar level at week-3 (VTmax 5.0 ± 1.6 vs. 2.7 ± 0.8, p = 0.029; SUVmax 1.9 ± 0.5 vs. 1.2 ± 0.2, p = 0.0012). VTmax correlated with SUVmax (R 2 = 0.532, p < 0.001). Conclusion: TSPO-PET imaging of immunocompetent SB28 mice facilitates early detection of tumor signals over sham lesions. SB28 tumors mirror high TSPO-PET signals of human glioblastoma and could serve as a valuable translational model to study TSPO as an imaging biomarker.

Novel App knock-in mouse model shows key features of amyloid pathology and reveals profound metabolic dysregulation of microglia.

BACKGROUND: Genetic mutations underlying familial Alzheimer's disease (AD) were identified decades ago, but the field is still in search of transformative therapies for patients. While mouse models based on overexpression of mutated transgenes have yielded key insights in mechanisms of disease, those models are subject to artifacts, including random genetic integration of the transgene, ectopic expression and non-physiological protein levels. The genetic engineering of novel mouse models using knock-in approaches addresses some of those limitations. With mounting evidence of the role played by microglia in AD, high-dimensional approaches to phenotype microglia in those models are critical to refine our understanding of the immune response in the brain. METHODS: We engineered a novel App knock-in mouse model (AppSAA) using homologous recombination to introduce three disease-causing coding mutations (Swedish, Arctic and Austrian) to the mouse App gene. Amyloid-ß pathology, neurodegeneration, glial responses, brain metabolism and behavioral phenotypes were characterized in heterozygous and homozygous AppSAA mice at different ages in brain and/ or biofluids. Wild type littermate mice were used as experimental controls. We used in situ imaging technologies to define the whole-brain distribution of amyloid plaques and compare it to other AD mouse models and human brain pathology. To further explore the microglial response to AD relevant pathology, we isolated microglia with fibrillar Aß content from the brain and performed transcriptomics and metabolomics analyses and in vivo brain imaging to measure energy metabolism and microglial response. Finally, we also characterized the mice in various behavioral assays. RESULTS: Leveraging multi-omics approaches, we discovered profound alteration of diverse lipids and metabolites as well as an exacerbated disease-associated transcriptomic response in microglia with high intracellular Aß content. The AppSAA knock-in mouse model recapitulates key pathological features of AD such as a progressive accumulation of parenchymal amyloid plaques and vascular amyloid deposits, altered astroglial and microglial responses and elevation of CSF markers of neurodegeneration. Those observations were associated with increased TSPO and FDG-PET brain signals and a hyperactivity phenotype as the animals aged. DISCUSSION: Our findings demonstrate that fibrillar Aß in microglia is associated with lipid dyshomeostasis consistent with lysosomal dysfunction and foam cell phenotypes as well as profound immuno-metabolic perturbations, opening new avenues to further investigate metabolic pathways at play in microglia responding to AD-relevant pathogenesis. The in-depth characterization of pathological hallmarks of AD in this novel and open-access mouse model should serve as a resource for the scientific community to investigate disease-relevant biology.

Reduced Acquisition Time [18F]GE-180 PET Scanning Protocol Replaces Gold-Standard Dynamic Acquisition in a Mouse Ischemic Stroke Model.

AIM: Understanding neuroinflammation after acute ischemic stroke is a crucial step on the way to an individualized post-stroke treatment. Microglia activation, an essential part of neuroinflammation, can be assessed using [18F]GE-180 18 kDa translocator protein positron emission tomography (TSPO-PET). However, the commonly used 60-90 min post-injection (p.i.) time window was not yet proven to be suitable for post-stroke neuroinflammation assessment. In this study, we compare semi-quantitative estimates derived from late time frames to quantitative estimates calculated using a full 0-90 min dynamic scan in a mouse photothrombotic stroke (PT) model. MATERIALS AND METHODS: Six mice after PT and six sham mice were included in the study. For a half of the mice, we acquired four serial 0-90 min scans per mouse (analysis cohort) and calculated standardized uptake value ratios (SUVRs; cerebellar reference) for the PT volume of interest (VOI) in five late 10 min time frames as well as distribution volume ratios (DVRs) for the same VOI. We compared late static 10 min SUVRs and the 60-90 min time frame of the analysis cohort to the corresponding DVRs by linear fitting. The other half of the animals received a static 60-90 min scan and was used as a validation cohort. We extrapolated DVRs by using the static 60-90 min p.i. time window, which were compared to the DVRs of the analysis cohort. RESULTS: We found high linear correlations between SUVRs and DVRs in the analysis cohort for all studied 10 min time frames, while the fits of the 60-70, 70-80, and 80-90 min p.i. time frames were the ones closest to the line of identity. For the 60-90 min time window, we observed an excellent linear correlation between SUVR and DVR regardless of the phenotype (PT vs. sham). The extrapolated DVRs of the validation cohort were not significantly different from the DVRs of the analysis group. CONCLUSION: Simplified quantification by a reference tissue ratio of the late 60-90 min p.i. [18F]GE-180 PET image can replace full quantification of a dynamic scan for assessment of microglial activation in the mouse PT model.

Loss of TREM2 rescues hyperactivation of microglia, but not lysosomal deficits and neurotoxicity in models of progranulin deficiency.

Haploinsufficiency of the progranulin (PGRN)-encoding gene (GRN) causes frontotemporal lobar degeneration (GRN-FTLD) and results in microglial hyperactivation, TREM2 activation, lysosomal dysfunction, and TDP-43 deposition. To understand the contribution of microglial hyperactivation to pathology, we used genetic and pharmacological approaches to suppress TREM2-dependent transition of microglia from a homeostatic to a disease-associated state. Trem2 deficiency in Grn KO mice reduced microglia hyperactivation. To explore antibody-mediated pharmacological modulation of TREM2-dependent microglial states, we identified antagonistic TREM2 antibodies. Treatment of macrophages from GRN-FTLD patients with these antibodies led to reduced TREM2 signaling due to its enhanced shedding. Furthermore, TREM2 antibody-treated PGRN-deficient microglia derived from human-induced pluripotent stem cells showed reduced microglial hyperactivation, TREM2 signaling, and phagocytic activity, but lysosomal dysfunction was not rescued. Similarly, lysosomal dysfunction, lipid dysregulation, and glucose hypometabolism of Grn KO mice were not rescued by TREM2 ablation. Synaptic loss and neurofilament light-chain (NfL) levels, a biomarker for neurodegeneration, were further elevated in the Grn/Trem2 KO cerebrospinal fluid (CSF). These findings suggest that TREM2-dependent microglia hyperactivation in models of GRN deficiency does not promote neurotoxicity, but rather neuroprotection.

Microglial activation states drive glucose uptake and FDG-PET alterations in neurodegenerative diseases.

2-Deoxy-2-[18F]fluoro-d-glucose positron emission tomography (FDG-PET) is widely used to study cerebral glucose metabolism. Here, we investigated whether the FDG-PET signal is directly influenced by microglial glucose uptake in mouse models and patients with neurodegenerative diseases. Using a recently developed approach for cell sorting after FDG injection, we found that, at cellular resolution, microglia displayed higher glucose uptake than neurons and astrocytes. Alterations in microglial glucose uptake were responsible for both the FDG-PET signal decrease in Trem2-deficient mice and the FDG-PET signal increase in mouse models for amyloidosis. Thus, opposite microglial activation states determine the differential FDG uptake. Consistently, 12 patients with Alzheimer's disease and 21 patients with four-repeat tauopathies also exhibited a positive association between glucose uptake and microglial activity as determined by 18F-GE-180 18-kDa translocator protein PET (TSPO-PET) in preserved brain regions, indicating that the cerebral glucose uptake in humans is also strongly influenced by microglial activity. Our findings suggest that microglia activation states are responsible for FDG-PET signal alterations in patients with neurodegenerative diseases and mouse models for amyloidosis. Microglial activation states should therefore be considered when performing FDG-PET.

Improving depth-of-interaction resolution in pixellated PET detectors using neural networks.

Parallax error is a common issue in high-resolution preclinical positron emission tomography (PET) scanners as well as in clinical scanners that have a long axial field of view (FOV), which increases estimation uncertainty of the annihilation position and therefore degrades the spatial resolution. A way to address this issue is depth-of-interaction (DOI) estimation. In this work we propose two machine learning-based algorithms, a dense and a convolutional neural network (NN), as well as a multiple linear regression (MLR)-based method to estimate DOI in depolished PET detector arrays with single-sided readout. The algorithms were tested on an 8× 8 array of 1.53× 1.53× 15 mm3 crystals and a 4× 4 array of 3.1× 3.1× 15 mm3 crystals, both made of Ce:LYSO scintillators and coupled to a 4× 4 array of 3× 3 mm3 silicon photomultipliers (SiPMs). Using the conventional linear DOI estimation method resulted in an average DOI resolution of 3.76 mm and 3.51 mm FWHM for the 8× 8 and the 4× 4 arrays, respectively. Application of MLR outperformed the conventional method with average DOI resolutions of 3.25 mm and 3.33 mm FWHM, respectively. Using the machine learning approaches further improved the DOI resolution, to an average DOI resolution of 2.99 mm and 3.14 mm FWHM, respectively, and additionally improved the uniformity of the DOI resolution in both arrays. Lastly, preliminary results obtained by using only a section of the crystal array for training showed that the NN-based methods could be used to reduce the number of calibration steps required for each detector array.