New Developments - Theranostics and Artificial Intelligence
Report template: SSTR PET/CT (Ga-68 DOTA peptides)
This chapter's English text is a draft translation (prepared with Claude and checked against a terminology glossary and automatic checks); specialist review is in progress. Please report errors with the feedback button.
Solve questions (27) →Contents
- 28.1 The Theranostic Concept and Applications
- 28.2 New Radiopharmaceuticals (FAPI, Fibroblast Activation Protein)
- 28.3 Artificial Intelligence and Deep Learning Applications
- 28.4 Total-Body PET/CT
- 28.5 Digital PET and New Detector Technologies
- Key Points for the Board
- Current guidelines on this chapter's topics
Learning objectives and modules3 modules · about 2 h 10 min
Learning objectives
After this chapter you will be able to:
- Explain the theranostic principle and the main theranostic pairs; apply patient selection criteria.
- Compare alpha and beta emitters in terms of range, LET and clinical use.
- Explain the targets and potential uses of new radiopharmaceuticals such as FAPI.
- Discuss the applications and limits of artificial intelligence and radiomics in nuclear medicine.
- Explain the advantages of total-body PET, digital PET and new detector technologies.
Modules
The chapter is divided into study modules without changing the order of the text: read, watch the animations, then answer the questions linked to those headings.
- Module 128.1 The Theranostic Concept and Applications30 min · 1 animation · 13 questionsAnswer the questions →
- Module 228.2–28.3 New Radiopharmaceuticals (FAPI, Fibroblast Activation Protein) · Artificial Intelligence and Deep Learning Applications55 min · 26 questionsAnswer the questions →
- Module 328.4–28.5 Total-Body PET/CT · Digital PET and New Detector Technologies · Key Points for the Board45 min · 2 animations · 16 questionsAnswer the questions →
Times are estimates: 130 words per minute, 3 min per animation, 1.5 min per question.
MOST FREQUENTLY ASKED IN THIS CHAPTER
✓
Theranostics: "See it → Treat it" → diagnosis + treatment directed at the same molecular target
✓
Theranostic pair: same ligand + different radionuclides (diagnostic + therapeutic)
✓
The most important theranostic pairs:
✓
NETTER-1: Lu-177 DOTATATE → midgut NET → PFS prolonged
✓
VISION: Lu-177 PSMA-617 → mCRPC → OS and rPFS prolonged
✓
Patient selection is critical: target expression must be confirmed by imaging
✓
NET – Krenning score ≥2 → suitable for PRRT
28.1 The Theranostic Concept and Applications13 questions
Definition and Basic Principles
•
Theranostics: "Therapy" + "Diagnostics" → combining diagnosis and treatment directed at the same molecular target
•
Basic principle: "See it → Treat it" → detect the target by imaging → apply treatment directed at the same target
•
The same ligand/peptide/antibody → labelled with different radionuclides:
•
Diagnostic radionuclide: gamma emitter (SPECT) or positron emitter (PET) → imaging
•
Therapeutic radionuclide: beta, alpha or Auger electron emitter → treatment
•
Examples of theranostic pairs:
–
Ga-68 / Lu-177 → DOTATATE (somatostatin receptor)
–
Ga-68 / Lu-177 → PSMA (prostate-specific membrane antigen)
–
I-123 / I-131 → NaI (sodium iodide – thyroid)
–
I-123 / I-131 → MIBG (adrenal medulla / neuroblastoma)
–
F-18 / Lu-177 → PSMA (DCFPyL / PSMA-617)
–
Cu-64 / Cu-67 → DOTATATE or PSMA
–
Ga-68 / Ac-225 → PSMA (alpha therapy)
BOARD TIP
▸
"What is a theranostic pair?"
→ Two radiopharmaceuticals binding to the same molecular target, one labelled with a diagnostic (PET/SPECT) and the other with a therapeutic radionuclide. The most classic examples: Ga-68/Lu-177 DOTATATE (NET) and Ga-68/Lu-177 PSMA (prostate cancer).
Figure 28.1: The theranostic principle: labelling a carrier that binds the same target with a diagnostic and a therapeutic radionuclide, and examples of clinical theranostic pairs (schematic).
Did this figure help you understand the topic?
Theranostic Applications – Clinical Examples
1. Neuroendocrine Tumours (NET) – Somatostatin Receptor Theranostics:
•
Target: somatostatin receptor type 2 (SSTR2)
•
Diagnosis: Ga-68 DOTATATE PET/CT → images SSTR2 expression
•
Treatment: Lu-177 DOTATATE (Lutathera®) → PRRT (Peptide Receptor Radionuclide Therapy)
•
NETTER-1 trial: Lu-177 DOTATATE → significantly prolonged PFS in midgut NETs
•
Patient selection: tumour uptake on Ga-68 DOTATATE PET/CT ≥ liver (Krenning score ≥2)
•
Treatment protocol: 4 cycles, 7.4 GBq (200 mCi) / cycle, 8 weeks apart
•
Renal protection: amino acid infusion (lysine + arginine) → reduces proximal tubular reabsorption
2. Prostate Cancer – PSMA Theranostics:
•
Target: Prostate-Specific Membrane Antigen (PSMA)
•
Diagnosis: Ga-68 PSMA-11 or F-18 DCFPyL (Pylarify®) PET/CT
•
Treatment: Lu-177 PSMA-617 (Pluvicto®)
•
VISION trial: Lu-177 PSMA-617 → significantly prolonged OS and rPFS in mCRPC (metastatic castration-resistant prostate cancer)
•
Patient selection: PSMA-positive metastatic disease on PSMA PET/CT + no PSMA-negative lesions (discordant disease → poor response)
•
Treatment protocol: 6 cycles, 7.4 GBq (200 mCi) / cycle, 6 weeks apart
•
Side effects: xerostomia (salivary gland uptake), bone marrow suppression, renal toxicity
BOARD TIP
▸
"What did the VISION trial show?"
→ Lu-177 PSMA-617 → prolonged OS and rPFS in mCRPC.
▸
"Patient selection criterion for PSMA therapy?"
→ Positive uptake on PSMA PET + no PSMA-negative lesions (discordant disease responds poorly to treatment).
▸
"The NETTER-1 trial?"
→ Lu-177 DOTATATE → prolonged PFS in midgut NETs.
3. Thyroid Cancer – Iodine Theranostics:
•
Target: sodium-iodide symporter (NIS)
•
Diagnosis: I-123 or low-dose I-131 whole-body scan
•
Treatment: high-dose I-131 (ablation / treatment of metastases)
•
The oldest and most classic theranostic application
•
Details: → See Chapter 13 (Endocrine System) and Chapter 26 (Radionuclide Therapy)
4. Neuroblastoma / Phaeochromocytoma – MIBG Theranostics:
•
Target: norepinephrine transporter (NET – uptake-1 mechanism)
•
Diagnosis: I-123 MIBG scintigraphy
•
Treatment: I-131 MIBG (Azedra® – high specific activity; production discontinued in 2024)
•
Details: → See Chapter 13 and Chapter 26
5. Radioimmunotherapy:
•
Target: CD20 (B-cell lymphoma)
•
Diagnosis: In-111 ibritumomab → biodistribution imaging
•
Treatment: Y-90 ibritumomab tiuxetan (Zevalin®)
•
Details: → See Chapter 20 and Chapter 26
Theranostic Radionuclides – Comparison
| Radionuclide | Type | T½ | Emission | Tissue Penetration | Use |
|---|---|---|---|---|---|
| Ga-68 | Diagnostic | 68 min | β⁺ (PET) | – | PET imaging |
| F-18 | Diagnostic | 110 min | β⁺ (PET) | – | PET imaging |
| Cu-64 | Diagnostic/Therapeutic | 12.7 hours | β⁺, β⁻ | ~1 mm (β⁻) | PET + therapeutic potential |
| Lu-177 | Therapeutic | 6.7 days | β⁻ (+ γ 208 keV) | ~2 mm (max) | PRRT, PSMA therapy |
| Y-90 | Therapeutic | 64 hours | β⁻ (pure) | ~11 mm (max) | Radioembolisation, RIT |
| Ac-225 | Therapeutic | 10 days | α | ~50-80 µm | Alpha therapy (TAT) |
| I-131 | Therapeutic | 8 days | β⁻ (+ γ 364 keV) | ~2 mm (max) | Thyroid, MIBG therapy |
| Bi-213 | Therapeutic | 46 min | α | ~50-80 µm | Alpha therapy (experimental) |
| Tb-161 | Therapeutic | 6.9 days | β⁻ + Auger | mean ~0.3 mm (max. ~2 mm) | Experimental (alternative to Lu-177) |
Alpha Therapy (Targeted Alpha Therapy – TAT)
•
Alpha particles: high LET (~80-100 keV/µm), short range (50-80 µm → a few cell diameters)
•
Advantages:
–
Very high cytotoxic effect → double-strand DNA breaks
–
Short range → minimal damage to surrounding tissue
–
Independent of the oxygen effect (OER ~1) → effective in hypoxic tumours too
•
Disadvantages:
–
Short range → may be insufficient in heterogeneous tumours
–
Daughter radionuclides (daughter products) are hard to control → may detach from the target through the recoil effect (see 2.1.1, "Why it matters: recoil")
–
Production and logistic difficulties
•
Ac-225 PSMA-617: promising results in mCRPC → response in patients refractory to Lu-177
•
Ra-223 dichloride (Xofigo®): approved alpha therapy for bone metastases → calcium mimetic → taken up in osteoblastic metastases
BOARD TIP
▸
"Advantage of alpha therapy over beta therapy?"
→ High LET → double-strand DNA breaks, short range → little damage to surrounding tissue, OER ~1 → effective in hypoxic tumours.
▸
"When is Ac-225 PSMA considered?"
→ In mCRPC patients refractory to Lu-177 PSMA.
Figure 28.2: Tissue range of theranostic radionuclides and the difference in cellular effect between alpha and beta emitters (schematic).
Did this figure help you understand the topic?
Dosimetry and Patient Selection
•
Imaging-based patient selection is critical in the theranostic approach:
–
Is target expression sufficient? → assess with PET/SPECT
–
Is the tumour/background ratio sufficient? → predicts treatment response
–
Are critical organ doses acceptable? → dosimetric planning
•
Krenning score (NET – DOTATATE):
•
1: uptake lower than liver
•
2: uptake equal to liver
•
3: uptake higher than liver
•
4: uptake equal to or higher than spleen
–
≥2 → suitable for PRRT
•
Assessment of PSMA expression:
–
All lesions should show PSMA uptake
–
PSMA-negative lesion → if FDG-positive → discordant disease → poor prognosis, low response to PSMA therapy
•
Pre-treatment dosimetry:
–
Critical organ doses are estimated by diagnostic imaging
•
Kidney dose: 23 Gy cumulative limit (PRRT)
•
Bone marrow dose: 2 Gy cumulative limit
Left: an SSTR PET schematic with liver lesions whose uptake rises, and PSMA and FDG PET schematics of the same patient. Right: the Krenning scale comparing lesion SUV with liver and spleen, and a lesion table with the VISION and TheraP selection criteria.
Continuous stream · no seeking
Radioligand therapy only treats lesions that show the target on imaging; a single lesion without the target (discordant) can decide the patient's outcome.
Source: Krenning score and discordance from the book (28.1); VISION selection criterion Sartor et al., NEJM 2021; TheraP criterion Hofman et al., Lancet 2021. Organ and lesion SUVs representative.
How to use: a single click on the drawing pauses and resumes; in the PSMA/FDG step ↑ ↓ or the buttons below change the scenario; use the time bar below to go to any moment; ← → change steps.
Did this animation help you understand the topic?
Figure 28.3: Imaging-based patient selection in theranostics: Krenning score and discordant disease on PSMA/FDG PET (schematic).
EANM – Theranostic Target and Dosimetry:
•
Ideal theranostic target: should be specific to the disease or tumour microenvironment, reachable from the circulation, and stably expressed over time and across metastases
•
The strongest evidence for personalised dosimetry is in liver radioembolisation; prospective evidence in other radioligand therapies is limited
•
Immunotheranostics: whole-body molecular imaging assesses the spatial and temporal heterogeneity of the tumour microenvironment; it has the potential to improve treatment selection and the efficacy of immunotherapy
•
For optimisation levels (L1-L3) and the Euratom BSS framework, see Chapter 26.12
•
Main developments approved or backed by phase 3 evidence in 2020-2026 (November 2026):
–
Therapy: approval of Lu-177 PSMA-617 before a taxane (March 2025) and in hormone-sensitive disease (July 2026); first-line PRRT with NETTER-2 (2024); Lu-177 DOTATATE approved from age 12 (2024) and its equivalent products (2026); Lu-177 edotreotide with COMPETE (2026); the first positive alpha phase 3 (Ac-225 DOTATATE, ACTION-1, 2026); Ra-223 + enzalutamide with PEACE-3 (with a bone-protecting agent); hepatic arterial PRRT (LUTIA: no increase in uptake shown) (→ Chapter 26)
–
Diagnosis: PSMA PET agents (Ga-68 PSMA-11, piflufolastat, gozetotide kits, flotufolastat); F-18 flurpiridaz PET MPI (2024; → Chapter 14); F-18 florquinitau (MK-6240) tau PET (2026; → Chapter 19); F-18 FET (Pixclara, 2026; → Chapter 21); exendin-4 (GLP-1R) PET for insulinoma (→ Chapter 13); Zr-89 girentuximab for clear cell RCC (approval pending; → Chapter 16)
28.2 New Radiopharmaceuticals (FAPI, Fibroblast Activation Protein)19 questions
FAPI (Fibroblast Activation Protein Inhibitor)
•
Target: Fibroblast Activation Protein (FAP)
•
FAP: a serine protease expressed on cancer-associated fibroblasts (CAF) in the tumour microenvironment
•
Very low expression in normal tissues → high tumour/background ratio
•
FAP expression: high in many solid tumours → pan-tumour target
•
Low physiological uptake in brain, liver and bone marrow → advantage over FDG
Ga-68 FAPI PET/CT:
•
Most commonly used: Ga-68 FAPI-46 and Ga-68 FAPI-04
•
Advantages:
–
Low physiological background → minimal brain, liver and bowel uptake
–
No patient preparation required (no fasting or blood glucose control → unlike FDG)
–
Imaging 10-60 minutes after injection (rapid protocol)
–
No problem in diabetic patients (with FDG, hyperglycaemia → false negatives)
•
Disadvantages:
–
Not tumour-specific → also taken up in wound healing, fibrosis, inflammation and degenerative changes
–
Large prospective studies are still limited
–
Lack of standardisation
Situations in Which FAPI Is Superior to FDG:
•
Peritoneal carcinomatosis → low sensitivity with FDG, high uptake with FAPI
•
Brain tumours → high cortical background with FDG, low background with FAPI
•
Hepatocellular carcinoma → variable uptake with FDG, high uptake with FAPI
•
Gastric / colorectal cancer → high uptake with FAPI
•
Pancreatic cancer → high uptake with FAPI, staging and treatment response
•
Cholangiocarcinoma → marked uptake with FAPI
BOARD TIP
▸
"Greatest advantage of FAPI over FDG?"
→ Low physiological background (especially brain and liver).
▸
"In which tumours is FAPI superior to FDG?"
→ Peritoneal carcinomatosis, brain tumours, HCC, gastric/colorectal cancer.
▸
"Is FAPI specific?"
→ NO → also taken up in inflammation, fibrosis and wound healing → not tumour-specific but stroma-specific.
Figure 28.4: Comparison of F-18 FDG and Ga-68 FAPI PET in terms of physiological distribution and tumour contrast (schematic).
Did this figure help you understand the topic?
Therapeutic Potential of FAPI (Theranostics):
•
Lu-177 FAPI or Y-90 FAPI → therapeutic use is being investigated
•
Ac-225 FAPI → alpha therapy potential
•
Problem: short tumour retention of FAPI ligands → limited therapeutic efficacy
•
New-generation FAPI ligands: longer tumour retention is targeted (FAPI-46 > FAPI-04)
SSTR Antagonists (EANM)
•
Antagonists also bind inactive SSTR2 and are less internalised; they therefore recognise more binding sites than agonists
•
Compared with agonists, contrast is higher, especially in the liver and spleen, and more lesions are detected
•
Ga-68 NODAGA-JR11: higher sensitivity than Ga-68 DOTATOC and DOTATATE, with a better target/background ratio in the liver
•
Lu-177 DOTA-JR11: tumour dose 1.7-10.6 times higher than that of the agonist; in a phase I/II trial the disease control rate was 94.7%, but grade 3-4 haematotoxicity was 42.5% (recommended: a total of 13 GBq over 3 cycles)
•
Second-generation antagonists: LM3 and LM4 (Ga-68 DATA5m-LM4, F-18 AlF-NOTA-LM3) show little liver uptake; with Lu-177 DOTA-LM3 a disease control rate of 85.1% was found, with no nephrotoxicity observed
•
Future: alpha-PRRT with Tb-149
GRPR (Gastrin-Releasing Peptide Receptor) Targeting (EANM)
•
GRPR: a G protein-coupled receptor of the bombesin family; activated by binding of GRP
•
Overexpression: prostate, breast, ovarian, uterine, lung, GIST and colon cancers; it is therefore a theranostic target usable in many tumours (pan-tumour)
•
The organ with the highest physiological expression is the pancreas (lower in the oesophagus and rectum)
•
GRPR antagonists are especially promising for imaging prostate and breast cancer
Other New PET Radiopharmaceuticals
1. F-18 PSMA-1007:
•
PSMA-targeted PET agent → alternative to Ga-68 PSMA
•
Advantages: F-18 → longer T½ (110 min vs 68 min), cyclotron production → central distribution possible, better spatial resolution
•
Disadvantage: bone uptake → false positives (benign bone lesions)
2. F-18 Fluciclovine (Axumin®):
•
Synthetic amino acid analogue → images amino acid transport
•
Indication: biochemical recurrence of prostate cancer
•
Its use is declining as PSMA PET becomes widespread
3. F-18 FET (Fluoroethyltyrosine):
•
Amino acid PET → in brain tumours
•
Low cortical background → superior to FDG
•
Tumour grade, treatment response, recurrence vs radionecrosis
4. F-18 FAZA / F-18 FMISO:
•
Hypoxia imaging → in radiotherapy planning
•
Hypoxic tumours → radioresistant → change of treatment strategy
5. C-11 Methionine:
•
Amino acid PET → brain tumours
•
Short T½ (20 min) → requires an on-site cyclotron
•
Valuable in distinguishing recurrence from radionecrosis
6. Ga-68 Pentixafor:
•
Target: CXCR4 (chemokine receptor)
•
In haematological malignancies (multiple myeloma, lymphoma)
•
Theranostic pair: Ga-68 Pentixafor / Lu-177 Pentixather
7. Cu-64-Labelled Agents:
•
Cu-64: T½ = 12.7 hours → emits β⁺ and β⁻ → potential for both imaging and therapy
•
Cu-64 DOTATATE: alternative to Ga-68 DOTATATE → longer T½ → delayed imaging possible
•
Theranostic pair: Cu-64 / Cu-67
8. Zr-89-Labelled Antibodies (Immuno-PET):
•
Zr-89: T½ = 78.4 hours → suited to the slow pharmacokinetics of antibodies
•
Zr-89 trastuzumab → imaging HER2 expression
•
Zr-89 atezolizumab → imaging PD-L1 expression
•
Potential in patient selection and prediction of treatment response
BOARD TIP
▸
"Why is Zr-89 used for immuno-PET?"
→ Antibodies have long half-lives (days) → the 78.4-hour T½ of Zr-89 matches antibody pharmacokinetics. F-18 or Ga-68 are far too short.
▸
"Property of Cu-64?"
→ Emits both β⁺ (PET) and β⁻ (therapy) → theranostic potential in a single radionuclide.
Immuno-PET and the Tumour Microenvironment (EANM)
•
Cancer-immunity cycle: antigen presentation, T-cell priming, tumour infiltration, recognition and killing; CTLA-4 regulates the early and PD-1/PD-L1 the late effector steps
•
Components of the tumour microenvironment: CAF, endothelium and pericytes, Treg, MDSC, TAM, dendritic cells, TIL; FAP-high CAFs are associated with immunosuppression and can be imaged with FAPI PET
•
CD8+ imaging shows the presence of cells; F-18 AraG (T-cell activation), granzyme B, IFN-γ and IL-2/IL-2R target effector function
•
FDG may also reflect immune cell glycolysis and pseudoprogression
•
CTLA-4, PD-1 and PD-L1 antibodies are mostly labelled with Zr-89 (delayed imaging required, logistically difficult); in future small molecules and peptides labelled with F-18 or Ga-68 will be more practical
Interventional Nuclear Medicine (iNM) (EANM)
•
The use of molecular imaging information for real-time or ex vivo guidance in biopsy, surgery and other interventions; its most established component is radioguided surgery (RGS)
•
Applications: SLN surgery, surgery with a target-directed tracer (e.g. PSMA-targeted surgery), image-guided biopsy, ROLL and RSL, theranostic liver radioembolisation
•
Because of its low energy the most suitable radionuclide is Tc-99m; with In-111 and PET radionuclides the staff dose increases
•
Hybrid tracers: the radioactive signal (in deep tissue) and fluorescence (high-resolution, real-time) are used together; e.g. ICG-Tc-99m-nanocolloid
•
Although the SLN detection rate approaches ~100%, the false-negative rate is 1-14%
•
ROLL: 2-15 MBq Tc-99m MAA is injected into the lesion; RSL: titanium seeds containing I-125 are used (long-term localisation)
•
De-escalation in breast cancer: MARI and targeted axillary dissection (TAD) reduce unnecessary ALND
•
Future: drop-in gamma and beta probes, robotic surgery, freehand SPECT, intraoperative PET/CT, augmented and virtual reality
28.3 Artificial Intelligence and Deep Learning Applications7 questions
Basic Concepts
Artificial Intelligence (AI) Hierarchy:
•
Artificial Intelligence (AI): machines exhibiting human-like intelligent behaviour → the broadest concept
•
Machine Learning (ML): algorithms that learn from data → a subset of AI
•
Conventional ML: requires feature engineering (manual feature extraction)
•
Algorithms: random forest, SVM (support vector machine), logistic regression
•
Deep Learning (DL): multilayer artificial neural networks → a subset of ML
–
Automatic feature extraction → learns directly from raw data
•
CNN (Convolutional Neural Network): the most common architecture for image analysis
•
RNN / LSTM: for sequential data (time series)
•
U-Net: a common architecture for medical image segmentation
•
GAN (Generative Adversarial Network): synthetic image generation
•
Transformer: natural language processing and, recently, image analysis
Figure 28.5: The hierarchy of artificial intelligence, machine learning and deep learning, and a comparison of conventional ML and deep learning workflows (schematic).
Did this figure help you understand the topic?
Areas of AI Application in Nuclear Medicine
1. Image Reconstruction and Enhancement:
•
Low-dose imaging: improving the quality of images acquired at low dose (low counts) with AI
–
Potential to reduce patient dose by 50-75% → contributes to the ALARA principle
–
Denoising → deep learning-based
•
Fast acquisition: enhancement with AI of images acquired in a short time
–
Shortening PET acquisition time → patient comfort, fewer motion artefacts
•
Attenuation correction: creating an attenuation map without CT (especially important in PET/MR)
–
Generating synthetic CT from MR images → attenuation correction
–
Direct estimation of the attenuation map from emission data
2. Automatic Segmentation:
•
Tumour segmentation: automatic calculation of metabolic tumour volume (MTV) and total lesion glycolysis (TLG) on FDG PET/CT
•
Organ segmentation: automatic delineation of critical organs for dosimetry
•
Bone segmentation: calculation of metastatic burden on bone scintigraphy (BSI – Bone Scan Index)
3. Diagnosis and Classification:
•
Myocardial perfusion scintigraphy: automatic detection of ischaemia/scar and risk classification
•
Bone scintigraphy: detection of metastases, benign/malignant differentiation
•
Lung V/Q scintigraphy: automatic assessment of PE probability
•
Dementia assessment: automatic pattern recognition on FDG PET and amyloid PET
•
Thyroid scintigraphy: nodule characterisation
4. Prognostic Assessment and Radiomics:
•
Radiomics: extracting high-dimensional quantitative features from medical images
•
First-order features: histogram-based (mean, standard deviation, skewness, kurtosis)
•
Second-order features: texture analysis (GLCM, GLRLM, GLSZM)
•
Higher-order features: filter-based (wavelet, Laplacian of Gaussian)
•
Radiomics + ML → radiomic signature:
–
Prediction of treatment response
–
Prognosis estimation
–
Assessment of tumour heterogeneity
–
Prediction of genetic mutations (radiogenomics)
•
Delta radiomics: change in radiomic features before and after treatment → early response assessment
BOARD TIP
▸
"What is radiomics?"
→ Extracting high-dimensional quantitative features from medical images → used for treatment response, prognosis and tumour characterisation.
▸
"What is a CNN?"
→ Convolutional Neural Network → the most common deep learning architecture for image analysis.
▸
"How does AI reduce dose in nuclear medicine?"
→ It improves low-count images by denoising → dose can be reduced by 50-75%.
Figure 28.6: Radiomics workflow: segmentation of the image, extraction of first-order/second-order/higher-order features and a radiomic signature with an ML model (schematic).
Did this figure help you understand the topic?
5. Treatment Planning and Dosimetry:
•
AI-based dosimetry in radionuclide therapy:
–
Automatic organ segmentation → dose calculation
–
Voxel-based dosimetry → more accurate dose distribution
–
Prediction of treatment response → personalised dose optimisation
•
In Y-90 radioembolisation: tumour/normal liver segmentation with AI → dose planning
Types of AI Learning and Data Issues (EANM)
•
Radiomics: automatic extraction of many quantitative features from images and building of diagnostic or predictive models with ML/DL
•
Conventional ML: hand-crafted features are used; prior knowledge can be built into the model, but feature selection can introduce bias. DL: less dependent on prior knowledge, but requires very large, high-quality data
•
Supervised learning (labelled data): random forest, SVM, k-nearest neighbours, logistic regression, boosting
•
Unsupervised learning (unlabelled data): clustering (hierarchical, k-means); dimensionality reduction (PCA, factor analysis, UMAP, t-SNE); denoising and data augmentation with autoencoders and GANs
•
On small datasets radiomic models are easier to train and interpret than DL; small, single-centre studies are prone to overfitting
•
Imbalanced data lead to poor performance in minority subgroups; FAIR data principle: findable, accessible, interoperable and reusable
•
"No free lunch" theorem: there is no universal method that is superior in every situation; several models should therefore be tried, with adequate cross-validation
•
Interpretability is the model's own transparency; explainability refers to methods that make a complex model understandable after the fact
•
When image enhancement methods are taken into clinical routine, clinical accuracy rather than quantitative quality metrics should be validated (e.g. 89% accuracy in amyloid PET with 1% of the dose)
•
DL: can generate pseudo-CT from MR alone, or attenuation- and scatter-corrected PET directly from uncorrected PET
Challenges and Limitations of AI
•
Data quality and quantity: deep learning requires large, high-quality, labelled datasets
•
Generalisability: a model trained at one centre → performance may fall at other centres
–
Different scanners, protocols, patient populations
•
Solution: multicentre data, federated learning
•
Explainability: the "black box" problem → the model cannot explain its decision
–
Explainability is important for clinical trust and regulatory approval
–
Explainability methods such as Grad-CAM and SHAP
•
Regulatory approval: FDA/CE approval process → AI software is "Software as a Medical Device (SaMD)"
•
Ethical issues: data privacy, algorithmic bias, accountability
•
Validation: prospective, multicentre clinical studies are needed
28.4 Total-Body PET/CT4 questions
Concept and Technology
•
Total-body PET/CT (TB-PET/CT): extending the axial FOV (field of view) to cover the whole body
•
Conventional PET/CT: axial FOV = 15-30 cm → requires multiple bed positions
•
Total-body PET/CT: axial FOV = ~194 cm (uEXPLORER) or ~106 cm (Biograph Vision Quadra) or ~142 cm (PennPET Explorer)
•
uEXPLORER: the first total-body PET/CT system → developed at UC Davis → axial FOV 194 cm
•
Biograph Vision Quadra (Siemens): axial FOV 106 cm → in clinical use
Advantages
1. Dramatic Increase in Sensitivity:
•
Geometric sensitivity: simultaneous photon collection from the whole body → ~40-fold increase in sensitivity (uEXPLORER vs conventional)
•
This increase in sensitivity can be used as follows:
•
Ultra-low-dose imaging: activity can be reduced to 1/40 → paediatric, pregnant, repeated scans
•
Ultra-fast imaging: acquisition time of seconds to minutes → fewer motion artefacts, greater patient comfort
•
Ultra-high-quality imaging: at standard dose and time → very high signal-to-noise ratio
•
Delayed imaging: sufficient counts even at low activity → imaging possible 4-8 hours later
2. Dynamic Whole-Body Imaging:
•
In conventional PET/CT, dynamic studies are possible only at a single bed position (limited FOV)
•
In TB-PET/CT, simultaneous dynamic acquisition over the whole body → kinetic analysis in all organs
•
Applications:
–
Whole-body Ki (influx constant) parametric maps → quantitative assessment of metabolic activity
–
Multi-organ pharmacokinetic modelling → drug development studies
–
Radiopharmaceutical biodistribution and dosimetry → whole-body dosimetry in a single session
–
Assessment of tumour heterogeneity → different kinetics in different lesions
3. New Clinical Applications:
•
Vascular imaging: simultaneous assessment of the entire vascular bed → atherosclerosis, vasculitis
•
Infection/inflammation: assessment of whole-body inflammatory burden → sarcoidosis, large-vessel vasculitis
•
Pharmacokinetic studies: real-time monitoring of drug distribution → drug development
•
Paediatric oncology: ultra-low dose → repeated scans become safe in children
•
Research: simultaneous study of physiological processes across the whole body
BOARD TIP
▸
"Most important advantage of total-body PET/CT?"
→ ~40-fold increase in sensitivity → this can be converted into ultra-low-dose, ultra-fast or ultra-high-quality imaging.
▸
"Axial FOV of uEXPLORER?"
→ 194 cm.
▸
"Axial FOV of conventional PET?"
→ 15-30 cm.
▸
"Why is dynamic whole-body imaging important?"
→ Simultaneous kinetic analysis in all organs → whole-body dosimetry and pharmacokinetics in a single session.
Left: annihilations in a patient seen from the side and photon pairs caught or lost depending on the axial FOV, then conventional and total-body PET images building up. Right: total-body and single-organ geometric efficiency of four systems relative to conventional, and how sensitivity, activity, time and decay set the counts.
Continuous stream · no seeking
The ~40-fold sensitivity of total-body PET comes both from seeing the whole body at once and from catching oblique photon pairs, and it can be spent on lower dose, shorter time or better quality.
Source: FOV lengths, ~40-fold gain and ways of use from the book (28.4). Geometric efficiency computed on this page by Monte Carlo (no attenuation, scatter or acceptance-angle limit); phantom and counts representative; F-18 T½ 109.8 min.
How to use: a single click on the drawing pauses and resumes; ↑ ↓ or the buttons below change the system's axial field of view in the first step and the way the sensitivity is used in the second; ← → change steps.
Did this animation help you understand the topic?
Figure 28.7: Total-body PET/CT: comparison of conventional and long axial FOV systems and the ways the ~40-fold increase in sensitivity can be used (schematic).
Limitations
•
Cost: very high equipment cost → limited access
•
Data management: huge data size → requires storage and processing capacity
•
Scatter and random coincidences: wide FOV → scatter fraction increases → correction algorithms needed
•
Attenuation correction: creating the CT attenuation map over a long axial FOV → truncation artefacts
•
Clinical evidence: large prospective studies still limited → cost-effectiveness not proven
28.5 Digital PET and New Detector Technologies12 questions
Conventional PET Detectors (Analogue)
•
Conventional PET: scintillation crystal + PMT (photomultiplier tube)
•
Crystal → converts the gamma photon into light → PMT → converts the light into an electrical signal
•
PMTs are large → limit detector design
•
Position determined by Anger logic → limited spatial resolution
•
Energy resolution: ~12-15% (at 511 keV)
•
Timing resolution: ~500-600 ps (for TOF)
Digital PET Detectors (SiPM)
SiPM (Silicon Photomultiplier):
•
A semiconductor photon detector replacing the PMT
•
Other names: MPPC (Multi-Pixel Photon Counter), dSiPM (digital SiPM)
•
Structure: an array of thousands of micro-APD (Avalanche Photodiode) pixels
•
Each pixel operates in Geiger mode → single-photon sensitivity
Advantages of Digital PET:
1. Superior timing resolution:
•
SiPM: ~200-215 ps (Biograph Vision) → ~60 ps (in new-generation prototypes)
•
PMT: ~500-600 ps
–
Better TOF performance → signal-to-noise ratio increases → detection of small lesions improves
2. Better energy resolution:
•
SiPM: ~9-11% (at 511 keV)
•
PMT: ~12-15%
–
Narrower energy window → scatter reduction → increased contrast
3. Higher spatial resolution:
–
SiPMs are small → 1:1 crystal-detector coupling is possible
–
Smaller crystal elements can be used (3 mm vs 4 mm)
•
Spatial resolution: ~3-4 mm (digital) vs ~4-5 mm (analogue)
4. Compact design:
–
SiPMs are very small → more compact detector ring
–
MR-compatible → can be used in PET/MR systems (unaffected by the magnetic field)
–
PMTs are affected by the magnetic field → cannot be used in PET/MR
5. High count rate capability:
–
Shorter dead time → little performance loss at high activity
–
Better performance with short-lived radionuclides (such as Rb-82)
BOARD TIP
▸
"What replaces the PMT in digital PET?"
→ SiPM (Silicon Photomultiplier).
▸
"Most important advantage of digital PET?"
→ Superior timing resolution (~200 ps vs ~500 ps) → better TOF performance.
▸
"Why are SiPMs used in PET/MR?"
→ Unaffected by the magnetic field (PMTs are affected).
▸
"Spatial resolution of digital PET?"
→ ~3-4 mm (analogue: ~4-5 mm).
TOF-PET (Time of Flight) – Advanced
•
TOF principle: the difference in arrival time of the two coincident photons at the detectors → narrows the position of the annihilation point along the LOR
Δx = c × Δt / 2
•
Δx: position uncertainty, c: speed of light, Δt: timing resolution
–
500 ps → Δx ≈ 7.5 cm
–
200 ps → Δx ≈ 3 cm
–
60 ps → Δx ≈ 0.9 cm (almost direct localisation)
•
Contribution of TOF to SNR: SNR gain ∝ √(D/Δx) (D: patient diameter)
–
The TOF advantage is more marked in large patients
•
The superior timing resolution of digital PET → dramatically increases TOF performance
In a PET ring: positron range, annihilation photons, coincidence and line of response (LOR), scattered, random and multiple events, and image formation (TOF).
Continuous stream · no seeking
Key point: when two 511 keV photons are detected within the timing window the event is placed on the line joining the two detectors; scatter and randoms create false lines, and TOF narrows the position along the line.
Source: book (mean positron ranges, 180° ± 0.25° deviation, T ∝ A, R ∝ τ·A²). Ring 80 cm, to scale; event-type proportions schematic.
How to use: a single click on the drawing pauses and resumes; in the range step ↑ ↓ or the rows of the list pick the radionuclide; in the image steps drag the lesion to move it, and with TOF on ↑ ↓ change the timing resolution; ← → change steps.
Did this animation help you understand the topic?
Figure 28.8: Relationship between timing resolution and position uncertainty in TOF-PET (Δx = c × Δt / 2) and the probability distribution along the LOR.
New Crystal Materials
| Crystal | Density (g/cm³) | Light Yield (photons/keV) | Decay Time (ns) | Use |
|---|---|---|---|---|
| BGO | 7.13 | 8-10 | 300 | Older PET systems |
| LSO | 7.40 | 20-30 | 40 | Modern PET |
| LYSO | 7.10 | 25-32 | 41 | Modern PET (most common) |
| LaBr₃ | 5.08 | 63 | 16 | Research (superior timing) |
| CeBr₃ | 5.10 | 60 | 19 | Research |
| NaI(Tl) | 3.67 | 38 | 230 | Gamma camera (SPECT) |
| CZT | 5.78 | – (direct conversion) | – | SPECT, CZT-SPECT |
CZT-SPECT (Cadmium Zinc Telluride)
•
CZT (CdZnTe): semiconductor detector → converts the gamma photon directly into an electrical signal
–
No intermediate scintillation crystal + PMT/SiPM → direct conversion
•
Advantages:
•
Superior energy resolution: ~5-6% (at 140 keV) vs NaI(Tl) ~9-10%
•
Compact design: small detectors → focused geometry
•
High sensitivity: focused collimator + CZT → 5-8-fold increase in sensitivity over a conventional gamma camera
•
Better spatial resolution: pixelated detector array
•
Clinical applications:
•
Cardiac SPECT: CZT-SPECT cardiac cameras (D-SPECT, Discovery NM530c)
•
Acquisition time: 2-5 minutes (conventional: 15-20 minutes)
•
Dose reduction: equivalent image quality at 50% of the dose
–
Dynamic SPECT → measurement of myocardial blood flow (MBF) possible (PET-like)
•
General-purpose SPECT: new-generation CZT-based general-purpose SPECT cameras
•
Thyroid imaging: high-resolution thyroid cameras
BOARD TIP
▸
"Most important advantage of the CZT detector?"
→ Direct conversion (gamma → electrical, no intermediate step) → superior energy resolution (~5-6%).
▸
"Advantage of the CZT-SPECT cardiac camera?"
→ 5-8-fold increase in sensitivity → acquisition time 2-5 minutes, dose can be reduced by 50%, MBF measurement possible with dynamic SPECT.
▸
"Why is CZT not used in PET?"
→ Insufficient stopping power at 511 keV → low density.
Figure 28.9: Signal chains of analogue (PMT), digital (SiPM) and CZT detectors, and comparison of the energy resolution of NaI(Tl) and CZT at 140 keV (schematic).
Did this figure help you understand the topic?
Other New Technologies
1. Photon-Counting CT:
•
Conventional CT: energy-integrating detectors → measure total energy
•
Photon-counting CT: counts each photon individually and measures its energy → spectral information
•
Advantages: better contrast resolution, dose reduction, material decomposition
•
Potential to improve attenuation correction quality in PET/CT and SPECT/CT
2. Monolithic Crystal Detectors:
•
A single-piece (monolithic) crystal instead of a pixelated crystal
•
Depth-of-interaction (DOI) information → parallax error decreases → resolution increases
•
At the research stage
3. Cherenkov Imaging:
•
Production of Cherenkov radiation in tissue by beta-emitting radionuclides (Y-90, F-18)
•
Imaging with an optical camera → potential intraoperative use
•
Verification of dose distribution after Y-90 radioembolisation
4. Compton Camera:
•
Collimator-free gamma imaging → electronic collimation instead of a mechanical collimator
•
Potential for very high sensitivity → no photon loss due to the collimator
•
Can operate over a wide energy range → different radionuclides
•
Not yet in clinical use → at the research stage
Key Points for the Board
The Theranostic Concept
•
Theranostics: "See it → Treat it" → diagnosis + treatment directed at the same molecular target
•
Theranostic pair: same ligand + different radionuclides (diagnostic + therapeutic)
•
The most important theranostic pairs:
–
Ga-68 / Lu-177 DOTATATE → NET (somatostatin receptor)
–
Ga-68 / Lu-177 PSMA-617 → mCRPC (prostate-specific membrane antigen)
–
I-123 / I-131 → thyroid cancer (NIS) and MIBG (norepinephrine transporter)
–
In-111 / Y-90 → radioimmunotherapy (CD20)
–
Cu-64 / Cu-67 → new-generation theranostic pair
–
Ga-68 / Ac-225 → alpha therapy (PSMA)
Clinical Trials
•
NETTER-1: Lu-177 DOTATATE → midgut NET → PFS prolonged
•
VISION: Lu-177 PSMA-617 → mCRPC → OS and rPFS prolonged
•
Patient selection is critical: target expression must be confirmed by imaging
Patient Selection Criteria
•
NET – Krenning score ≥2 → suitable for PRRT
•
Prostate – PSMA PET positive + no discordant disease → suitable for PSMA therapy
•
Discordant disease: PSMA-negative / FDG-positive → poor prognosis, low response to PSMA therapy
•
Cumulative kidney dose limit: 23 Gy (PRRT)
•
Cumulative bone marrow dose limit: 2 Gy
•
Renal protection: amino acid infusion (lysine + arginine)
Alpha Therapy (TAT)
•
High LET (~80-100 keV/µm) → double-strand DNA breaks
•
Short range (50-80 µm) → minimal damage to surrounding tissue
•
OER ~1 → effective in hypoxic tumours too
•
Ac-225 PSMA: promising in mCRPC refractory to Lu-177
•
Ra-223: approved alpha therapy → osteoblastic bone metastases
•
Disadvantage: recoil effect of daughter radionuclides → detachment from the target
FAPI PET/CT
•
Target: FAP (Fibroblast Activation Protein) → cancer-associated fibroblasts
•
Most important advantage: low physiological background (brain, liver)
•
Superior to FDG: peritoneal carcinomatosis, brain tumours, HCC, gastric/colorectal cancer
•
No patient preparation required (no fasting or blood glucose control)
•
Not tumour-specific → also taken up in wound healing, fibrosis and inflammation
•
Therapeutic potential: Lu-177 FAPI, Ac-225 FAPI → at the research stage
•
Problem: short tumour retention → limited therapeutic efficacy
New Radiopharmaceuticals
•
F-18 PSMA-1007: alternative to Ga-68 PSMA → longer T½, central distribution → disadvantage: bone false positives
•
Zr-89: T½ = 78.4 hours → immuno-PET → suited to antibody pharmacokinetics
•
Cu-64: T½ = 12.7 hours → both β⁺ and β⁻ → theranostics in a single radionuclide
•
Ga-68 Pentixafor: CXCR4-targeted → haematological malignancies
Artificial Intelligence
•
CNN: the most common deep learning architecture for image analysis
•
U-Net: medical image segmentation
•
Dose reduction: AI denoising of low-count images → 50-75% dose reduction
•
Radiomics: extracting quantitative features from images
•
1st order: histogram (mean, SD, skewness)
•
2nd order: texture analysis (GLCM, GLRLM)
•
Higher order: filter-based (wavelet, LoG)
•
Challenges: data quality, generalisability, explainability ("black box"), regulatory approval
Total-Body PET/CT
•
uEXPLORER: axial FOV = 194 cm → the first total-body PET/CT
•
Biograph Vision Quadra: axial FOV = 106 cm
•
Conventional PET: axial FOV = 15-30 cm
•
Increase in sensitivity: ~40-fold (uEXPLORER)
•
Options for use: ultra-low dose OR ultra-fast OR ultra-high quality
•
Dynamic whole-body imaging: simultaneous kinetic analysis in all organs → dosimetry in a single session
Digital PET and New Detectors
•
SiPM: a semiconductor photon detector replacing the PMT
•
Timing resolution: ~200 ps (PMT: ~500 ps)
•
Energy resolution: ~9-11% (PMT: ~12-15%)
–
Unaffected by the magnetic field → used in PET/MR
–
Compact design → small size
•
TOF formula: Δx = c × Δt / 2
–
500 ps → 7.5 cm | 200 ps → 3 cm | 60 ps → 0.9 cm
–
SNR gain ∝ √(D/Δx) → the TOF advantage is more marked in large patients
•
CZT: direct-conversion detector (gamma → electrical, no intermediate step)
•
Energy resolution: ~5-6% (NaI(Tl): ~9-10%)
•
Cardiac CZT-SPECT: 5-8-fold increase in sensitivity → 2-5 min acquisition, 50% dose reduction
–
Dynamic SPECT → MBF measurement possible (PET-like)
–
Insufficient stopping power at 511 keV → not used in PET
•
Crystal comparison: LYSO (most common in modern PET) > LSO > BGO (older) | NaI(Tl) → SPECT
•
Photon-counting CT: counts each photon individually → spectral information → material decomposition
•
Compton camera: collimator-free → potential for very high sensitivity → not yet in clinical use.
Additional Notes from EANM Guidelines:
•
SSTR antagonists (JR11, LM3, LM4): less internalisation, higher contrast in the liver; haematotoxicity with JR11
•
GRPR: pan-tumour target (prostate, breast); highest physiological expression in the pancreas
•
Immuno-PET: checkpoint antibodies with Zr-89; F-18 AraG for T-cell activation; FAP-high CAF → immunosuppression
•
iNM: hybrid tracer ICG-Tc-99m-nanocolloid; SLN false negatives 1-14%
•
AI: radiomics has the advantage on small data; 'no free lunch' → several models; interpretability ≠ explainability
Current guidelines on this chapter's topics
The latest guidelines and consensus documents on the procedures covered in this chapter (official EANM index and primary sources, November 2026). Comparison of the text with these versions is in progress; consult the guideline itself for details.
Finished the chapter?
This chapter has 27 questions. Solve them now to consolidate the topic.
You can also swipe sideways to change chapter