Artificial intelligence (AI) has vast potential to reshape nuclear medicine. Applications can be found at every step of the processing workflow, including image acquisition, image reconstruction, enhancement, and registration, segmentation, extraction of image-derived biomarkers, and prognostication, both for clinical and preclinical research and routine use. Emerging perspectives include computational nuclear oncology, foundation models, biomorphic AI, and quantum AI - currently exploratory but rapidly evolving. However, clinical translation remains limited. This narrative review summarizes the current literature on the subject and highlights current developments and future research directions. As such, it is not intended as a systematic analysis; instead, it outlines potential future directions, supported by multiple examples. Analysis of literature reveals impressive advances in the application of AI to nuclear medicine, such as improved time resolution, robust denoising, and automated lesion segmentation. Key challenges include the need for high-quality reference data, generalizability across scanners and populations, transparency and uncertainty quantification of model decisions, and compliance with evolving regulatory frameworks. For the future, progress will require broad collaboration between AI developers from research and industry, clinicians, hospital information technology, and regulators, while structured initiatives such as joint symposia should actively include patients to ensure innovations address real needs. Education across disciplines will also be critical for building trust and competence towards fully embracing the potential of AI for better (nuclear) medicine.

Aim/Introduction: Distance-encoding biomorphic-informational neural network (DEBI-NN) is a recently proposed architecture in which connection weights are defined by the distances between neurons positioned in a Euclidian space. This approach drastically reduces the number of trainable parameters compared to classical neural networks in which weights are directly trained. The training process for DEBI-NN is based on a genetic algorithm (GA), rather than gradient descent (GD) which remains the prevailing optimization algorithm in deep learning. We aim to design and implement a GD learner for DEBI-NN and assess its performance compared to GA. Materials and Methods: We designed a spatial backpropagation scheme tailored to DEBI-NN and carried out a comparison between GD and GA for classification tasks, using a synthetic non-linear "two-moons" dataset, two clinical medical imaging radiomic datasets and a fetal cardiotocography dataset with a sample sizes ranging from n=85 to n=2126. Each optimizer was tuned through targeted hyperparameter searches adapted to each dataset. Results: Across all experiments, GA consistently produced superior decision boundaries and classification performance (Synthetic: 100 Conclusion: These findings highlight fundamental limitations of gradient-based methods in architectures with highly interdependent spatial parameters and confirm the suitability of evolutionary strategies for training DEBI-NN.

Positron emission tomography (PET) with long axial field-of-view (LAFOV) scanners generates unprecedentedly large datasets, posing major challenges for conventional reconstruction algorithms. Recent advances in deep learning have improved PET image quality, yet scaling to high-dimensional data remains computationally demanding. Quantum computing (QC) offers new opportunities to address such challenges. This work presents a pilot study on image denoising using a hybrid quantum-classical autoencoder. The model employs a classical encoder, a quantum bottleneck implemented via a parameterized quantum circuit, and a classical decoder. Using simulated experiments on the medical modified national institute of standards and technology (MedMNIST) dataset with 28 × 28 grayscale images and on PET scan images with 64 × 64, results demonstrated that the quantum bottleneck can effectively recover structural details from noisy inputs. While scaling to large-volume PET data will require advances in quantum hardware, circuit design, and error mitigation, these results highlight the potential of hybrid quantum–classical approaches to complement deep learning in PET reconstruction.

Objective: Reliable identification of fibrotic regions is essential for targeted catheter ablation therapy, as current imaging modalities such as cardiac magnetic resonance imaging face technical and clinical limitations, particularly in resolution and compatibility with implanted devices. This work presents the quantitative assessment of optical coherence tomography (OCT) images to classify myocardium into fibro-elastic versus normal. Methods: We acquired ultrahigh resolution OCT images from a sheep model with chronic myocardial infarction and performed pixelwise depth-resolved analysis to generate attenuation coefficient maps. In addition, we extracted radiomic features from three dimensional subvolumes to train a XGBoost classifier and validated our results against histological ground truth using Masson's trichrome staining histology to assess diagnostic accuracy. Results: Attenuation and prediction probabilities effectively highlighted fibro-elastic regions. Widefield en face representations offered fast three dimensional screening of cardiac fibrosis. The radiomics-based XGBoost classifier achieved an area under the curve of 0.97 for binary classification. Conclusions: Combining ultrahigh resolution OCT with a straightforward attenuation coefficient and a robust radiomics pipeline for optical property extraction and high throughput radiomic feature analysis enables label-free assessment of fibrotic microstructures in the myocardium. Significance: The proposed quantitative framework enhances the detection and characterization of fibrotic myocardial tissue, offering potential for improved diagnostic precision and clinical integration of OCT in cardiology workflows towards data-driven catheter therapy guidance.

Nuclear medicine (NM) has progressed from planar imaging to 3D/4D hybrid modalities, yet advances in artificial intelligence (AI) for analysing imaging and non-imaging NM data have not been widely adopted in clinical routine. NM typically operates with small, heterogeneous, multi-centric datasets – conditions that challenge conventional AI. EARL acquisition standards and IBSI radiomic guidelines were established to mitigate the issue, while radiomic and image-level harmonization methods have been proposed to address multicentric differences. Nevertheless, generalization remains limited in NM AI models. Deep learning (DL) models can generalize when trained on very large datasets, which NM rarely has. Consequently, to date, DL had been mostly utilized in NM for segmentation tasks. As of 2026, vision transformers, large language models, and foundation models are promising efforts, albeit their utilization in NM remains limited. We consider that NM can benefit from novel AI capable of achieving highly-accurate predictive performance and generalization with dramatically fewer trainable parameters as well as training samples compared to prior solutions. We introduce quantum computing (QC) as a potential solution and discuss its properties that address data scarcity on the AI side. We highlight present and near-term opportunities enabled by quantum AI and propose resources to support engagement with and adoption of QC in NM, with the goal of advancing patient care.

Artificial Intelligence (AI) approaches in clinical science require extensive data preprocessing (DP) steps prior to building AI models. Establishing DP pipelines is a non-trivial task, mainly driven by purely mathematical rules and done by data scientists. Nevertheless, clinician presence shall be paramount at this step. The study proposes a data preprocessing approach driven by clinical domain knowledge, where clinician input, in form of explicit and non-explicit rules, directly impacts the algorithms’ decision-making processes, thus, making the DP planning phase more inclusive for clinicians. The rule set table (RST) was introduced as interface which accepts clinician’s input as formal rules (including four actions: exp-keep, exp-remove, pref-keep, pref-remove features or samples) in human-readable form and translates it to machine readable input for preprocessing algorithms. A collection of commonly used algorithms was incorporated for data preprocessing of various clinical cohorts in both single and multi-center scenarios. The impact of RST was evaluated by utilizing 100-fold Monte Carlo cross-validation scheme for prostate and glioma cohorts (single center) with 80 − 20

Introduction To date, small, imbalanced datasets are considered challenging to efficiently train deep learning (DL) models, especially in the medical domain. Consistently, most Artificial Intelligence (AI) approaches in conjunction with small datasets rely on shallow radiomics where traditional machine learning (ML) is utilized for analysing image-derived features. In this study, we evaluate a recently introduced spatial neural network scheme called Distance-Encoding Biomorphic-Informational Neural Network (DEBI-NN), which trains spatial coordinates of artificial neuron coordinates instead of weights, that are then calculated from neuron distances. This technique dramatically reduces the number of parameters to train, thereby making DEBI-NN eligible for the analysis of small, imbalanced datasets. We refer to this property as spatial plasticity. We hypothesized that DEBI-NNs could systematically outperform baseline NN models in small clinical datasets while requiring less regularizations to be implemented, as spatial plasticity may have self-regularization properties. To test our hypothesis, we aimed to compare DEBI-NNs with baseline NNs while relying on various regularization techniques to investigate how DEBI-NNs perform in the presence of regularizers in small multi-centric medical imaging datasets. Methods Three multi-centric datasets were collected including diffuse large B-cell lymphoma (DLBCL) [18F]FDG positron emission tomography (PET)/computed tomography (CT) with clinical parameters to predict 2-years event-free survival; the head and neck [18F]FDG PET/CT dataset from the 2022 MICCAI challenge (HECKTOR), predicting human papillomavirus status; and [68Ga]Ga-PSMA-11 (PSMA-11) PET/CT as well as PSMA-11 PET/magnetic resonance imaging (MRI) cases to predict histopathology-provided International Society of Urological Pathology (ISUP) grades as low (ISUP ≤2) and high (ISUP >2) risk. Per cohort, 5 different network configurations having 1, 2 and 3 hidden layers and neuron count configurations were defined. Per configuration, DEBI-NNs had 7 regularization techniques and baseline NNs had 6 regularization configurations, totalling 27 = 128 and 26 = 64 regularization variants per network scheme to train and evaluate. Test balanced accuracy (BACC) was measured for each model and correlation of the test BACC in the presence of regularization techniques was evaluated in DEBI-NN and baseline NN models. Results The best-performing DEBI-NN models yielded 84.5 %, 80 % and 80.5 % BACC in DLBCL, HECKTOR and PROSTATE datasets, respectively. In contrast, the highest-performing baseline NN models yielded 71.9 %, 77.3 % and 77.3 % BACC in the same cohorts, respectively. In addition, baseline NNs required the implementation of more regularization techniques to increase test BACC from an average test BACC of 53 % (no regularization) to 60 % (6 regularizations), while DEBI-NNs needed no regularization to achieve 62 % BACC. In return, DEBI-NN BACC monotonously fell down to 56 % BACC as the number of regularizations increased. Conclusions DEBI-NNs exhibit a significantly simpler training complexity compared to baseline NNs, while they also outperform baseline NNs with the presence of minimal or no regularization techniques. Our results strongly imply that DEBI-NNs have a potential to pave the way for the utilization of neural networks in small and imbalanced medical datasets, which the field of medical imaging research routinely operates with.

Myocardial infarction, a leading cause of mortality worldwide, leaves survivors at significant risk of recurrence caused by scar-related re-entrant ventricular tachyarrhythmias. Effective treatment with ablation therapy requires a precise guidance system. Non-linear optical microscopy techniques, such as second harmonic generation (SHG) and two-photon excited fluorescence (TPEF), are promising candidates for a high-resolution alternative to conventional electrical mapping for assessing infarcted cardiac tissue. Here, we apply SHG and TPEF with a resolution advantage over commonly used electrical mapping techniques to assess ex-vivo sheep heart infarction. Analyzing conventional and radiomic features allows for quantitative characterization of scar tissue. Our machine learning classifier achieved high accuracy, offering a promising, data-driven approach for guiding in-situ ablation therapy with increased precision. This study represents a significant step towards integrating quantitative image analysis in therapeutic interventions.

Predicting blood-brain barrier (BBB) penetration is crucial for developing central nervous system (CNS) drugs, representing a significant hurdle in successful clinical phase I studies. One of the most valuable properties for this prediction is the polar surface area (PSA). However, molecular structures are missing geometric optimization, which, together with lack of standardization, leads to variations in calculation. Additionally, prediction rules have been established by combining different molecular properties such as the BBB score or CNS multiparameter optimization (CNS MPO). This study aims to create an approach for 3D PSA calculation, to directly apply this value in combination with a set of 23 other parameters in a novel machine learning (ML)-based scoring, and to further evaluate existing prediction models using a standardized database. We developed and analyzed a standardized data set derived from the same laboratory, encompassing 24 calculated and experimentally determined molecular parameters such as PSA from various models, HPLC log P values, and hydrogen bond characteristics for 154 radiolabeled molecules and licensed or well-characterized drugs. These molecules were classified into categories based on BBB penetration, nonpenetration, and interactions with efflux transporters. We supplemented these with a novel in silico 3D calculation of nonclassical PSA. Additionally, we have calculated published prediction rules based on this standardized and transparent database. Using these data, we trained various ML models within a 100-fold Monte Carlo cross-validation framework to derive a novel ML-based prediction score for BBB penetration and validated the three most used existing predictive rules. To interpret the influence of individual molecular parameters and different existing predictive rules, we employed explainable artificial intelligence methods including Shapley additive explanations (SHAP) and surrogate modeling. The ML approach outperformed existing scores for BBB penetration by applying a complex nonlinear integration of molecular properties, with the random forest model achieving the best performance for predicting binary BBB penetration (area under the receiver operating characteristic curve (AUC) 0.88, 95% confidence intervals: 0.87-0.90), and multiclass efflux transporter versus CNS-positive and CNS-negative prediction (AUC 0.82, 95% CI: 0.81-0.82). SHAP analysis revealed the multifactorial nature of the problem, highlighting the advantage of multivariate models over single predictive parameters. The ML model's superior predictive capability was demonstrated in comparison with existing scoring systems, like the CNS MPO (AUC 0.53), the CNS MPO Positron emission tomography (PET) (AUC 0.51), and BBB score (AUC 0.68) while also enabling the identification of efflux transporter substrates and inhibitors. Our integrated ML approach, combining experimental and in silico measurements with novel in silico methods based on a standardized database including a plethora of different substance groups (licensed drugs and in vivo evaluated PET tracers), enhances the prediction of BBB penetration. This approach may reduce the reliance on extensive experimental measurements and animal testing, accelerating CNS drug development.

Key objectives of precision medicine include improving treatment efficacy, reducing side effects, preventing disease, optimizing healthcare resources and costs, and ensuring that treatments are cost-effective while ultimately enhancing overall health outcomes. It matches therapies to patients most likely to benefit, accounts for individual differences in drug metabolism and responses, identifies individuals at higher risk, and implements preventive measures. Digital twins, creating virtual computational replicas of physical systems, processes, or entities, offer a unique opportunity to support personalized medicine through real-time simulation, monitoring, and optimization. In this roadmap, we propose recommendations on the workflow and methodology, describing how AI can be applied within image-based digital twin systems in nuclear computational oncology. We unite interdisciplinary expertise to efficiently guide the research and drive computational nuclear oncology toward clinical applications. We provide definitions, key insights, and a clear framework for each step in the development of a virtual system. These include: 1) Data Extraction and Digital Twin Creation, 2) Tracer Diffusion Simulation, 3) Scaling to Macroscale Tumor Analysis, and 4) Dynamic Updates and Continuous Learning. We discuss both clinical and technical perspectives, challenges, and promising future directions to bridge the gap between “real systems” and “virtual replicas”.

Understanding the process of fibrotic scarring of the myocardium is critical for the diagnosis and risk stratification of life-threatening cardiac dysfunction. Complex changes in structure, composition, and conductivity occurring at different stages of fibrogenesis diversify the biomedical characteristics of the myocardium. We present a multimodal optical imaging approach including cardiac optical mapping (COM), optical coherence tomography (OCT), multiphoton microscopy (MPM), and line scan Raman microspectroscopy (LSRM) for multiparametric assessment of the myocardium with radiomic analysis to link electrophysiologic, morphologic, functional, and molecular changes in ischemic cardiac tissue and validate our results with histology. COM is used to map the electrical behavior across myocardial tissue. Second harmonic generation and two-photon excitation fluorescence imaging as MPM techniques provide additional unique contrast of collagen, the extracellular matrix, and cardiac cells, such as cardiomyocytes playing a critical role in cardiac fibrosis. Our machine learning model based on radiomic features extracted from MPM data addresses the need for automated fast high-throughput classification between healthy and pathologic cardiac tissues and achieved an accuracy of 0.99. In addition, LSRM assesses the molecular contrast and is used to evaluate the development stage of fibrotic scarring and multiclass classification by utilizing partial least-squares discriminant analysis, achieving sensitivity and specificity values of 0.94. OCT is used for fast navigation through the sample, for intermodal referencing, and easy coregistration between the complementary imaging techniques operating at different fields of view and resolutions ranging from cm2 down to μm2.

In medicine, digital twins (DTs) serve as computational models that replicate biological and physiological characteristics of a specific individual — whether a patient, an organ, or even a single cell — and simulate virtual biomedical experiments. DT-based simulations hold the potential to identify the most beneficial intervention at any given moment. A range of technological approaches has been explored across various medical fields, with oncology being one of the most suitable areas of application in view of the necessity of timely and personalized treatment decisions. Medical imaging, and especially nuclear medicine, might have a central role in the development of DTs. In this review we define digital twins, examine current evidence, and discuss opportunities that digital twins can offer in computational nuclear oncology. We also briefly summarize the state-of-the-art of DTs in other fields.

Purpose: This study aims to assess whole-mount Gleason grading (GG) in prostate cancer (PCa) accurately using a multiomics machine learning (ML) model and to compare its performance with biopsy-proven GG (bxGG) assessment. Materials and Methods: A total of 146 patients with PCa recruited in a pilot study of a prospective clinical trial (NCT02659527) were retrospectively included in the side study, all of whom underwent 68Ga-PSMA-11 integrated positron emission tomography (PET) / magnetic resonance (MR) before radical prostatectomy (RP) between May 2014 and April 2020. To establish a multiomics ML model, we quantified PET radiomics features, pathway-level genomics features from whole exome sequencing, and pathomics features derived from immunohistochemical staining of 11 biomarkers. Based on the multiomics dataset, five ML models were established and validated using 100-fold Monte Carlo cross-validation. Results: Among five ML models, the random forest (RF) model performed best in terms of the area under the curve (AUC). Compared to bxGG assessment alone, the RF model was superior in terms of AUC (0.87 vs 0.75), specificity (0.72 vs 0.61), positive predictive value (0.79 vs 0.75), and accuracy (0.78 vs 0.77) and showed slightly decreased sensitivity (0.83 vs 0.89) and negative predictive value (0.80 vs 0.81). Among the feature categories, bxGG was identified as the most important feature, followed by pathomics, clinical, radiomics and genomics features. The three important individual features were bxGG, PSA staining and one intensity-related radiomics feature. Conclusion: The findings demonstrate a superior assessment of the developed multiomics-based ML model in whole-mount GG compared to the current clinical baseline of bxGG. This enables personalized patient management by identifying high-risk PCa patients for RP.

You have accessJournal of UrologySurgical Technology & Simulation: Artificial Intelligence III (PD36)1 May 2024PD36-06 CHARACTERIZING PROSTATE CANCER RISK WITH 68GA-PSMA-11 PET IMAGING AND AI Marcus Hacker, Simon Wail, Lubos Dolezel, David Iommi, Thomas Beyer, and Laszlo Papp Marcus HackerMarcus Hacker , Simon WailSimon Wail , Lubos DolezelLubos Dolezel , David IommiDavid Iommi , Thomas BeyerThomas Beyer , and Laszlo PappLaszlo Papp View All Author Informationhttps://doi.org/10.1097/01.JU.0001008916.72488.6a.06AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Prostate cancer (PC) is one of the most common types of cancer in men. Routine diagnosis of PC relies on image-guided biopsy sampling. However, biopsy-based risk estimations are accurate in only 60-70% cases. In this study we demonstrate the feasibility of providing non-invasive imaging-based PC risk predictions through artificial intelligence (AI) approaches. METHODS: A total of 50 patients who underwent 68Ga-PSMA-11 PET/CT imaging and image-guided biopsy sampling were included. Each CT image was automatically delineated by an nnU-Net deep learning (DL) model to detect the prostate. The resulting prostate mask was utilized to identify regions in the corresponding PET image in which another nnU-Net DL model identified suspicious lesions to generate a whole prostate lesion probability map (P-MAP). This P-MAP together with the corresponding PET image were analyzed to extract fuzzy radiomic features that conform to the Imaging Biomarker Standardization Initiative (IBSI). Each case was labelled based on their invasive biopsy Gleason Scores (GS) to distinguish high (GS >= 4+3) versus low (GS <= 3+4) risk cases. A supervised, automated machine learning approach (Dedicaid Ltd, Austria) was used to build 100 mixed-stacked ensemble learners relying on the radiomic features and their corresponding high-low risk labels. These 100 model instances were merged to compose a super-learner model (SLM).Another 68Ga-PSMA-11 PET/MRI cohort (n=24) served as an independent test set for estimating the performance of the SLM. Here, the reference labels were derived from whole-mount histopathology. Independent test performance was estimated by confusion matrix analytics. Following the evaluation with SLM, sensitivity (SNS), specificity (SPC), positive and negative predictive values (PPV and NVP) and area under the receiver operator characteristics curve (AUC) were calculated for the test cases. RESULTS: Independent SLM testing yielded 88% SNS, 82% SPC, 88% PPV, 82% NPV, and 89% AUC. CONCLUSIONS: This study demonstrates that DL in combination with super learners that employ standardized IBSI fuzzy radiomics accurately characterize high-vs-low risk PC in 68Ga-PSMA-11 PET imaging with a significantly higher predictive performance compared to standard clinical processes involving invasive tissue sampling. Source of Funding: Telix Pharmaceuticals © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e794 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Marcus Hacker More articles by this author Simon Wail More articles by this author Lubos Dolezel More articles by this author David Iommi More articles by this author Thomas Beyer More articles by this author Laszlo Papp More articles by this author Expand All Advertisement PDF downloadLoading ...

ObjectivesRadical prostatectomy (RP) is a common intervention in patients with localized prostate cancer (PCa), with nerve-sparing RP recommended to reduce adverse effects on patient quality of life. Accurate pre-operative detection of extraprostatic extension (EPE) remains challenging, often leading to the application of suboptimal treatment. The aim of this study was to enhance pre-operative EPE detection through multimodal data integration using explainable machine learning (ML).MethodsPatients with newly diagnosed PCa who underwent [68Ga]Ga-PSMA-11 PET/MRI and subsequent RP were recruited retrospectively from two time ranges for training, cross-validation, and independent validation. The presence of EPE was measured from post-surgical histopathology and predicted using ML and pre-operative parameters, including PET/MRI-derived features, blood-based markers, histology-derived parameters, and demographic parameters. ML models were subsequently compared with conventional PET/MRI-based image readings.ResultsThe study involved 107 patients, 59 (55%) of whom were affected by EPE according to postoperative findings for the initial training and cross-validation. The ML models demonstrated superior diagnostic performance over conventional PET/MRI image readings, with the explainable boosting machine model achieving an AUC of 0.88 (95% CI 0.87-0.89) during cross-validation and an AUC of 0.88 (95% CI 0.75-0.97) during independent validation. The ML approach integrating invasive features demonstrated better predictive capabilities for EPE compared to visual clinical read-outs (Cross-validation AUC 0.88 versus 0.71, p = 0.02).ConclusionML based on routinely acquired clinical data can significantly improve the pre-operative detection of EPE in PCa patients, potentially enabling more accurate clinical staging and decision-making, thereby improving patient outcomes.Critical relevance statementThis study demonstrates that integrating multimodal data with machine learning significantly improves the pre-operative detection of extraprostatic extension in prostate cancer patients, outperforming conventional imaging methods and potentially leading to more accurate clinical staging and better treatment decisions.Key PointsExtraprostatic extension is an important indicator guiding treatment approaches.Current assessment of extraprostatic extension is difficult and lacks accuracy.Machine learning improves detection of extraprostatic extension using PSMA-PET/MRI and histopathology.