Advancements in radiation dose reduction for pediatric CT head Imaging: A scoping review of emerging Technologies, Protocols, and optimization strategies

Highlights:

• •Pediatric patients are more radiosensitive than adults, making radiation dose optimization crucial in CT.
• •Traditional methods by modifying scanning parameters have limitations, due to the trade-off between dose and image quality.
• •Deep learning image reconstruction offers balance between dose reduction and image quality.

Abstract

Computed tomography (CT) is an important imaging modality that provides cross-sectional images, aiding in the detailed visualization of internal structures for accurate diagnosis and treatment. The pediatric population is more sensitive to radiation than adults, making radiation dose (RD) optimization an important concern in pediatric CT imaging. This scoping review emphasizes advanced RD reduction methods used in pediatric CT head imaging for diagnosing various clinical conditions with optimum RD and diagnostic image quality (IQ). A detailed search was conducted across five databases, such as PubMed, Scopus, CINAHL, Web of Science (WOS) and Embase using relevant keywords. A total of 24 articles were included in the final review. RD parameters and IQ related data were extracted from each article. Conventional RD reduction techniques in CT such as reducing the tube voltage, tube current and other scanning parameters, face limitations particularly in the pediatric population. These techniques lead to a trade-off between a lower RD and poor IQ which might obfuscate diagnostic details due to decreased contrast resolution with greater image noise and artifacts. To balance RD and diagnostic IQ, advanced technologies such as iterative reconstruction (IR) and deep learning image reconstruction (DLIR) algorithms with ultra-low dose protocols are increasingly being used. Hence, the review concludes that, compared with conventional dose reduction techniques, artificial intelligence based DLIR algorithms enhance IQ even for ultra-low dose protocols across various clinical domains in pediatric CT imaging.

Introduction

In recent years, the use of pediatric computed tomography (CT) head imaging has been increasing rapidly because of advancements in CT technologies which aid in the accurate diagnosis and treatment of various diseases. However, among diagnostic imaging procedures, CT examinations account for the highest proportion of patient radiation dose (RD) [1], [2]. Compared with adults, children are significantly more vulnerable to radiation-induced cancer [3], [4]. The lifetime attributable risk (LAR) of cancer from a single head CT examination could be ten times greater for an infant than for an adult [5]. Moreover, individuals with conditions necessitating multiple CT examinations receive cumulative RDs that amplify the potential risk [6].

Pediatric CT dose optimization could be more challenging than that for adults as drastic changes in body size occur within age groups [[31], [32]]. However, as recommended by the International Commission on Radiological Protection (ICRP) 135, for pediatric CT head examinations, protocols specific to age (instead of weight) can be used [33]. An important consideration for any imaging procedure is to obtain diagnostically acceptable image quality (IQ) sufficient for clinical purposes. Various techniques can be used in CT for optimizing the RD but decreasing the exposure factors can result in poor IQ resulting in a suboptimal study. RD for pediatric CT examination can be optimized by choosing appropriate CT acquisition parameters such as pitch, peak tube voltage (kVp), tube current exposure time product (mAs), rotation time, slice thickness, and increment particularly with respect to age and body size [[34], [35]]. Most studies have developed low dose (LD) protocols using phantoms to balance RD and IQ and then validated them in patients [35].

Despite the availability of various RD reduction methods for CT, existing reviews and guidelines are predominantly focused on adults and emphasize that conventional strategies offer limited guidelines for pediatric practice. With advancements in CT technologies, protocols and optimization strategies there is currently no comprehensive synthesis that evaluates their applicability and effectiveness in pediatric CT imaging. Addressing this gap is essential for informing evidence- based practice, guiding future research and supporting the development of pediatric specific recommendations. Thus, this review aims to synthesize current RD reduction strategies in pediatric CT head imaging, evaluate approaches that balance dose optimization with diagnostic IQ, and outline emerging advancements in pediatric dose reduction.

Methods

Search strategy

The current scoping review was performed as per the Preferred Reporting Item for Systematic Review and Meta-analysis extension for Scoping Reviews (PRISMA −ScR) guidelines [36]. PubMed, Scopus, CINAHL, Web of Science (WOS) and Embase were searched with keywords: “Radiation Dose”, “Optimization”, “Pediatric population”, “Computed Tomography”, and “Head examination” to find the relevant studies published through 2024 (Supplementary material 1: Search strategy). Boolean operators such as “AND” and “OR” were used. Two independent reviewers screened the titles and abstracts of the included studies, with disagreements resolved through discussion or third reviewer consultation. Only articles published in English were included in the review. Mendeley was used as the reference manager.

Eligibility criteria

We included both phantom and clinical studies to provide a comprehensive overview of RD reduction strategies, capturing methodological developments from phantom investigations and their translation into clinical practice for pediatric CT head imaging. These strategies include tube voltage reduction, tube current–time product modulation, automatic exposure control, pitch and collimation adjustments, scan length and time optimization, and the use of iterative reconstruction (IR) and deep learning image reconstruction (DLIR) algorithms. The exclusion criteria were articles that were conference abstracts, case reports, case series, letters, editorials, reviews or meta-analysis and published in languages other than English.

Data extraction

One reviewer performed the data extraction and charting independently and verified the data with the second reviewer. The following data were extracted from each article: author, publication year, sample size, details of the CT system, image acquisition parameters, dose optimization technique, image reconstruction algorithm, radiation dose [volume CT dose index (CTDI_vol), dose length product (DLP), effective dose (E)], percentage of dose reduction and IQ.

Results

Study selection

A comprehensive search across five databases (PubMed, Scopus, Web of Science, CINAHL, and Embase) initially identified 220 records. After 131 duplicates were removed, 89 records remained for title and abstract screening. Of these, 61 were excluded because of exclusion criteria. The full texts of 28 articles were then assessed for eligibility, and 4 were excluded because of insufficient data or inappropriate comparison groups. Ultimately, 24 studies were included in the final review, as illustrated in the PRISMA-ScR flow diagram (Fig. 1).

Fig. 1.

PRISMA- ScR flow chart for study selection.

Study characteristics

The review included a total of twenty-four studies, consisting of twenty studies involving human participants (5084 individuals) and four phantom studies. These studies were conducted across various countries, including Belgium (n = 1), Cameroon (n = 1), China (n = 2), Egypt (n = 1), Finland (n = 1), France (n = 1), Greece (n = 1), India (n = 3), Israel (n = 1), Korea (n = 4), Malaysia (n = 1), Poland (n = 1), Taiwan (n = 1) and the USA (n = 5). These studies used CT scanners manufactured by General Electric (GE) Healthcare (64, 128, 256 and 512 slices), Philips Healthcare (64,128 and 256 slices), Siemens Healthineers (64 and 128 slices), Neusoft medical system (64 slices) and Canon medical system (64 slices). Nine studies followed prospective methods, 13 studies followed retrospective methods, and two studies followed both methods: retrospective methods for standard dose (SD) and prospective methods for LD data collection. The detailed characteristics of the studies included in the review are shown in Table 1.

Table 1.

Characteristics of the studies included in the review.

Author, year and country	Study design	Reconstruction technique	Slice CT/ Vendor	Sample size	Dose reduction technique	Image quality parameters	Outcome of the study
Udayasankar et al., 2008 [7]USA	Retrospective	FBP	4-channel MDCT, LightSpeed QXi; GE Healthcare	SD: 92LD: 92	SD: 120 kVp,220 mAsLD: 120kVp tube voltage, 80 mAs	Quantitative analysis: Noise, Attenuation, GM conspicuity, CNR Qualitative analysis: IN, diagnostic acceptability, GM-WMD, sharpness of subarachnoid space margins, visualization of posterior fossa structures, SA, need for further imaging	LD CT showed 63 % reduction in RD with higher IN. However, IQ was diagnostically acceptable for the assessment of ventricular volume and shunt patency for follow-up evaluation in patients with hydrocephalus.
Jończyk-Potoczna et al., 2012 [8]Poland	Retrospective	−	128 slice, SOMATOM Definition AS, Siemens	SD: 128LD: 128	SD For BM < 20 kg: CareDose 4D (190 mAs), 120 kV For body mass ≥ 20 kg: 350 mAs, 120 kV LD For BM < 20 kg: CareDose 4D (100 mAs), 120 kV For BM ≥ 20 kg: 200 mAs, 120 kV	−	The lower mAs protocol reduced DLP by 40 % while maintaining diagnostic acceptable IQ for evaluation of VP shunt failure with hydrocephalus
Morton et al, 2013 [9]USA	Retrospective	−	−	SD: 2150LD: 624	SD: 120 kVp, 124–240mAs LD 1 (Half) : 120 kVp, 62––120 mAsLD 2 (quarter): 120 kVp, 30 – 42.5 mAs	Qualitative analysis:Information needed for the evaluation of hydrocephalus, skull, facial bones, ICH and satisfaction of OIQ	The indicated based LD mAs protocol resulted in reduction of RD without affecting IQ and diagnostic utility.
Vorona et al., 2013 [10] USA	Retrospective Prospective	FBP, ASIR 20 %	64 slice, LightSpeed VCT, HighSpeed CT/I unit	SD: 16LD: 16	SD: 120kVp, 120 mAs, FBP LD: 120 kVp, 96 mAs, 20 % ASIR	Quantitative analysis: Noise in air, WM, bone and CSF.Qualitative analysis: Diagnostic Acceptability, Sharpness, noise, artifacts	The use of 20 % ASIR resulted in 22 % reduction in RD
McKnight et al. 2014 [11] USA	Retrospective	ASIR	64 slice VCT GE healthcare	SD: 49LD:33	SD: 120 kVp,100 mAs (0–3 years), 180 mAs(3–12 years),260 mAs (>12 years); reconstruction mode: FBP LD: 120 kVP; 100 mAs (0–3 years), 125 mAs (3–12 years),150 mAs (>12 years); reconstruction mode: ASIR	Quantitative analysis: IN and attenuation Qualitative analysis: GWMD, sharpness, and OIQ	With the use OF ASIR 30 % 28–48 % reduction in CTDI vol was achieved for pediatric CT head examination
Kaasalainen et al. 2015 [12] Finland	Prospective	ASIR 30 %, ASIR 50 %Veo	64 slice, Discovery 750 HD, GE Health care	Phantom study	Newborn phantom: 80kVp and 100 kVp, fixed 10 mA with tube current modulation with noise index 35–55 5-year phantom: 80kVp and 100 kVp with fixed 10 mA and noise index 35120kVp with noise index: 35	Quantitative analysis: IN and contrast Qualitative analysis: Noise and artifacts	The use of ultra-low dose protocol with MBIR resulted in 83 % and 88 % reduction in mean organ dose for newborn and 5-year phantom respectively with adequate image quality.
Morel et al., 2016 [13]France	Prospective	SAFIRE – level 3	64- slice MDCT, Somatom AS+, Siemens AG	Group 1: 20Group 2: 20	SD: 120 kVp tube voltage, ATCM (400 mAs), 1 s rotation time, 0.6 pitchLD: ATVS (80 kVp – 120 kVp) ATCM (400 mAs), 1 s rotation time, 0.6 pitch	Quantitative analysis: Supratentorial and infratentorial contrast, noise, CNR Qualitative analysis: GM-WMD Delimitation of: peri-mesencephalic CSF space, shape of ventricularSystem, Visualization of the BG	With the activation of ATVS, 100kVp tube potential was selected for group 2 patients and resulted in 21 % reduction in CTDIvol with no significant difference in CNR and higher GM-WMD
Ernst et al. 2016 [14]Belgium	Retrospective	FBP, MBIR − VEO	SD: 64 slice Brillance CT, Philips Health careLD: 64 slice, Discovery 750 HD, GE Health care	SD: 24LD: 24	SD: 120 kVp, 276 mA, reconstruction mode: FBP LD: 80kVp, 10 mA, reconstruction mode:	Quantitative analysis: SNR and sharpness Qualitative analysis:IN, sharpness, OIQ and artefacts	The LD protocol showed 97 % reduction in CTDIv with similar OIQ, objective noise, sharpness, artifacts as SD. The LD protocol also showed higher objective sharpness and reduced subjective IN
Park et al., 2017 [15]South Korea	Retrospective	SAFIRE- strength 2.	128-slice,Dual source multidetector row, Somatom Definition Flash, Siemens Healthcare	SD:30LD: 30	SD: 120 kVp tube voltage, ATCM (220 mAs), 0.28 s rotation time, 0.6 pitchLD: 80 kVp tube voltage, ATCM (700mAs) , 0.5 s rotation time, 0.5 pitch	Quantitative analysis: attenuation, IN, SNR, CNR, PFAIQualitative analysis: IN, GM-WMD, artifact, OIQ	The LD protocol along with IRs showed improved GM-WM CNR and similar overall IQ as SD with 6.5 % reduction in RD
Kim et al. 2017 [16] Korea	RetrospectiveProspective	FBP, ASIR V	Revolution EVO, GE Healthcare	SD:38 LD: 28	SD: <3 years: 80kVp, 200mAs ≥3 – 15 years: 100kVp, 200 mAs LD <3 years: 80kVp, 180mAs, 30 % ASIR-V ≥3 – 15 years: 100kVp, 140 mAs, 40 % ASIR −V	Quantitative analysis: Noise, CNR Qualitative analysis:GWMD, sharpness, overall diagnostic quality	The use of age-based LD protocol along with ASIR −V resulted in 12.8 % to 34 % reduction in RD with improved or comparable IQ to SD.
Chang et al, 2017[17]Taiwan	Prospective	FBP, iDose4-level 3	128 slice, Brilliance iCT, Philips Health care	Phantom study	Tube voltage (kVp): 80,100, 120, tube current–time(mAs): 50, 100,150, 200, 250, 300, 350, and 400.	Quantitative analysis: accuracy of CT number, noise, SNR and CNR, resolution, FOM	IR technique can reduce radiation dose with 40 % increase in CNR. 80kVp/50 mAs are found to be optimal for pediatric CT head
Papadakis et al., 2019 [18]Greece	Prospective	FBP, ASIR 40 %	64-detector CT scanner, GE Medical Systems	Phantom study	Protocol A: ATCM and ATVS deactivated Protocol B: ATCM was activatedProtocol C: ATCM and ATVS activated	Quantitative analysis: attenuation, IN, CNR	For pediatric CT head, the use of ATCM reduced CTDIvol (8 % − 24 %) without affecting IN. But use of ATVS increased radiation dose up to 21 %
Vidyasagar et al., 2019 [19]India	Retrospective	−	Philips brilliance 6 and ingenuity core 128 slice.	SD: 43LD: 27	SD: 350 mAs/120kVpLD: 250mAs/100kVp	−	LD protocol with reduced mAs, kVp, FOV and increased pitch resulted in significant dose reduction
Southard et al. 2019 [20]USA	Retrospective	FBPIMR	SD: 64 slice, Brillance, Philips Healthcare LD: 256 slice, Brillance iCT, Philips Healthcare	SD:173LD:190	SD:120 kV mAs:200, 225,300 or 350; reconstruction mode: FBP LD: 120Kv; mAs:160, 168, 200 or 210; reconstruction mode: IMR	Quantitative analysis: SNR and CNR Qualitative analysis: anatomical details and G-WMD	LD protocol with IMR reconstruction resulted in 12–27 % reduction in radiation dose with improved SNR, CNR and subjective IQ
Sherif et al. 2020 [21]Egypt	Retrospective	FBP, ASIR V	Revolution EVO, GE Healthcare	SD:14LD:27	SD: 120 kV, 148 mA, rotation: 0.6 s, pitch 0.531:1, reconstruction: FBP LD: 120 kV, 40–100 mA, rotation: 0.6 s, pitch 0.531:1, reconstruction: ASIR-V 60 % and 80 %	Quantitative analysis: SNR, CNR Qualitative analysis:GWMD, sharpness, visibility of the ventricular system, and OIQ	ASIR-V algorithm resulted in significant reduction in RD with acceptable IQ (ED = 1.04 ± 0.1 mSv for LD versus 3.48 ± 0.45 mSv for SD)
Cho et al. 2020 [22]South Korea	Prospective	FBP, ADMIRE	128 −slice dual source CT, Somatom definition flash, Siemens Healthcare	SD:105LD: 109	SD:110 kVp, 65 mA (< 1 year), 85 mA (1–12 years), 110 mA (>12 years); reconstruction mode: FBP LD: 110 kVp, 50 mA (< 1 year), 55 mA (1–12 years), or 80 mA (>12 years); reconstruction mode: ADMIRE	Quantitative analysis: noise, SNR and CNR Qualitative analysis: noise, G-WMD, sharpness, artifacts, OIQ	Use of LD protocol along with ADMIRE resulted in 30.6 %, 32.1 % and 32.1 % reduction in CTDIv, DLP and ED with improved IQ
Sun et al., 2021[23]China	Retrospective	FBP 50 % ASIR V 100 % ASIR VDLIR (High)	256-row, Revolution CT, GE Healthcare	50	Tube voltage (kVp):120,Tube current(mA): 150 (0–2 years), 170 (3–6 years) 190 (7–12 years) 210 (≥13 years)	Quantitative analysis: SNR and CNRQualitative analysis: clarity of sulci/cisterns, boundaries between GM-WM, OIQ	The 5 mm DLIR −high images showed lowest IN and 0.625 mm DLIR images showed similar IN (3.11 ± 0.58 HU) and overall IQ score (3.04 ± 0.33) as 5 mm 50 % ASIR −V images (3.16 ± 0.60 HU and 3.05 ± 0.23). However, 0.625 mm DLIR- high images showed higher lesion detection (69 lesions) compared to 5 mm 50 % ASIR-V images (65 images).
Bingyang et al., 2021[24]China	Prospective	−	64-slice,Light Speed, GE Healthcare	SD: 15LD: 68	SD: 120 kVp, 250 mAs LD1: For HC 54.1–57.0 cm, 120 kVp, 200 mAs LD 2: For HC 51.1–54.0 cm, 120 kVp 150 mAs LD 3: For HC 48.1–51.0 cm, 120 kVp, 100 mAs	Quantitative analysis: attenuation, noise, SNR and CNR Qualitative analysis:noise, artifacts, anatomical details and lesions	The HC based low mAs protocol for CT head examination resulted in significant reduction in organic specific (brain, eye lenses and salivary glands)
Eddy et al., 2021 [25] Cameroon	Retrospective	IR	64 slice CT, Neusoft	145	100–120 kV, ATCM (149–400 mA)	−	For pediatric CT head examination, limiting the scan length to second c-spine based on indication is the effective way to reduce RD.
Atri et al. 2021 India [26]	Prospective	FBP, MBIR	SD: 64 slice Toshiba aquilionLD: 256 slice, Brillance iCT	SD: 41LD: 47	SD: 5–10 years: 120 kVp, 180 mAs 10–16 years: 120 kVp, 260 mAs LD: 5–10 years: 120 kVp, 100 mAs 10–16 years: 120 kVp, 150 mAs	Quantitative analysis: Noise, CNR, SNR Qualitative analysis: Noise, sharpness, GMWMD, Diagnostic acceptability,artefacts	The LD protocol with MBIR resulted in 79.8 %, 88.5 %, 81 % reduction in DLP, CTDIv, ED respectively with-out significant difference in IQ, IN and diagnostic acceptability.
Muhammad et al., 2022 [27]Malaysia	Prospective	iDose4	128 slice, Ingenuity core, Philips Healthcare	Phantom study	Tube voltage (kVp):100, Tube current–time, mAs: 402, 359, 320, 286 and 255	Quantitative analysis: SNR and CNR Qualitative analysis: sharpness, IN, artifacts	IR technique with reduced mAs resulted in 30 % reduction in radiation dose with improved SNR AND CNR.
Rabinowich et al. 2023 [28]Israel	Retrospective	FBPIMR	256-slice Brillance iCT Philips Healthcare	148	Tube voltage: 120 kV; Tube current (mA):91 (<1 year), 99 (1–5 years),144(5–10 years) and 180 (> 10 years);	Quantitative analysis: SNR and CNRQualitative analysis: G-WMD, OIQ, artifacts, visibility of anatomical structures	IMR group showed higher SNR, CNR and qualitative score compared to FBP group.
Lee et al., 2023 Korea [29]	Retrospective	FBP ASiR-V – 30 % and 40 %DLIR – low, medium and high	512-slice, Revolution CT, GE Healthcare	126	< 1 year: tube voltage: 100 kVp, reference mA: 250, 1–6 year: tube voltage: 100 kVp, reference mA: 290 6–12 years: tube voltage: 120 kVp, reference mA: 200 >12 years: tube voltage: 120 kVp, reference mA: 270	Quantitative analysis: SNR and CNR Qualitative analysis: noise, GM-WM differentiation, sharpness, artifact, acceptability, unfamiliar texture change	DLIR can significantly reduce IN with increased SNR, CNR, GM-WM differentiation, sharpness and provided better IQ compared with ASIR-V
Priyanka et al., 2024[30]India	Prospective	iDose4 – level 3	128-slice Incisive CT, Philips Healthcare	SD: 143LD: 71	SD: <1 year:100 kV, 200mAs 1–5 years: 120 kV/ 250mAs LD: <1 year:80 kV, 150mAs 1–5 years: 100 kV, 200mAs	Quantitative analysis: attenuation, IN, SNR, CNR, FOMQualitative analysis: IN, GM-WMD, artifact, OIQ	LD protocol by reducing kV and mAs and reconstructing the images with iDose4-level-3 can provide ultra-low RD with diagnostically acceptable IQ

ADMIRE: Advanced modeled iterative reconstruction, ASIR: Adaptive statistical iterative reconstruction, ATCM: Automatic tube current modulation, ATVS: Automatic tube voltage selection, BG: Basal ganglia, BM: body mass, CNR: Contrast to noise ratio, CSF: cerebrospinal fluid, CT: Computed Tomography, CTDIvol: volumetric computed tomography dose index, DLIR Deep learning image reconstruction, DLP: Dose length product, ED: effective dose, FBP: Filter back projection, FOM: Figure of merit, GM-WMD: Grey matter-white matter differentiation, HC: Head circumference, ICH: Intracerebellar hemorrhage, IMR: Iterative model reconstruction, IN: Image noise, IQ: Image quality, IR: Iterative Reconstruction, kVp: kilovoltage peak, LD: Low dose, mAs: milli amperage second, MBIR: Model-based Iterative Reconstruction MDCT: Multidetector row computed tomography, OIQ: Overall image quality PFAI: Posterior fossa artifact index, RD: Radiation dose, SA: Streak artifact, SAFIRE: Sinogram affirmed iterative reconstruction, SD: Standard dose, SNR: Signal to noise ratio, VP: Ventriculo-peritoneal shunt failure.

RD reduction by reducing scanning parameters

Reducing RD while maintaining IQ is crucial for pediatric CT examinations. Several RD reduction techniques such as reduced tube voltage, tube current-exposure time product, scan length, scan time, pitch and collimation have been employed by various studies and found to be effective. The RD indices such as CTDIvol, DLP, ED and percentage reduction in RD reported by various studies are shown in Table 2.

Table 2.

Radiation dose parameters.

year	Age group	CTDIv (mGy)		DLP (mGy.cm)		ED (mSv)		% RD reduction
		SD	LD	SD	LD	SD	LD
asankar et al., 2008 [7]	8 months – 21 years	43.4	15.5	695.1	252.8	1.6	0.58	63.4
Jończyk-Potoczna et al., [8]	1 month – 18 years	−	−	933	563	−	−	40
Vorona et al., 2013 [10]	3–18 years	28.8	22.4	444.5	338.4	−	−	23.9
McKnight et al., 2014 [11]	0–18 years	29.7	21.8	−	−	−	−	−
Morel et al., 2016 [13]	< 1 year	22.92	18.18	−	−	−	−	21
Ernst et al.,2016 [14]	0–35 months	32.18	0.94	487.84	15.04	3.07	0.08	97
Park et al., 2017 [15]	6 months – 15 years	24.7	23.6	479.8	448.9	1.31	1.23	4.7 – 6.9
Kim et al.,2017 [16]	0–15 years	18.06	14.80	276.47	215.74	−	−	10.3–34
Vidyasagar et al., 2019 [19]	0–15 years	−	−	919	439	−	−	−
Southard et al., 2019 [20]	0–18 years	29.9 – 33.3	21.8 – 29.2	−	−	−	−	12–27
Sherif et al.,2020 [21]	3–10 years	35.61	14.36	614.23	256.7	3.48	1.04	75.6
Cho et al., 2020 [22]	0–18 years	12.4 – 26.2	8.9––17.7	155.6––400.7	106.7–––280.8	0.84–0.76	0.58 – 0.52	30–––32
Sun et al., 2021[23]	0.1–14 years	−	18.18	−	269.43	−	−	−
Atri et al., 2021[26]	5 – 16 years	85.86	9.85	656.86	132.59	1.84	0.35	79.8 – 88.5
Rabinowich et al., 2023 [28]	0–16 years	−	15.68	−	328.79	−	1.5	−
Lee et al., 2023[ 35]	4 – 204 months	−	10.5 – 17.7	−	135.1––297.5	−	0.73––0.42	−
Priyanka et al., 2024[30]	0 – 5 years	16.86––27.54	6.67 – 15.18	453.75 – 687.03	137.91––287.70	2.9 – 2.78	0.91–1.16	44.88–––68.62

CTDIvol: volumetric computed tomography dose index, DLP: Dose Length Product, ED: effective dose, LD: low dose, mSv: milli sievert, mGy: milli. Gray, mGycm: milli gray centimeter, RD: Radiation dose, SD: standard dose.

Lowering the tube voltage from 120 kV to 80 kV has resulted in a significant reduction in RD while maintaining or improving IQ when it is used in conjunction with the IR technique compared with the conventional filtered back projection (FBP) technique [[20], [21]]. Priyanka et al [30] demonstrated RD reduction of up to 68 % for children under 1 year of age and a 58 % reduction in RD for the 1–5 years age group using iDose⁴ reconstruction and the low kV technique.

The automatic tube voltage and tube current selection on the basis of body composition can also optimize RD, as reported by Papadakis et al., with RD reduction of up to 70 %[18]. Furthermore, lowering the tube current (mA) reduces RD but also increases image noise which may be handled using IR approaches, as reported by Udayashankar et al (63 % RD reduction)[7]. The scanning protocols tailored to the body mass or head circumference can further reduce RD while maintaining diagnostic IQ. Additionally, adjusting the scan length and time along with the usage of a higher pitch can also contribute to reducing the RD. However, usage of a higher pitch affects the IQ if not combined with IR techniques[25].

Advanced image reconstruction techniques

Advanced image reconstruction approaches including advanced model iterative reconstruction (ADMIRE), adaptive statistical iterative reconstruction (ASIR), model-based iterative reconstruction (MBIR), iterative model reconstruction (IMR), iDose⁴, and DLIR have shown great promise in improving CT imaging [37], [38], [39]. These strategies can significantly minimize RD while maintaining or even increasing IQ for pediatric CT examinations. MBIR paired with ultra-low-dose methods has been demonstrated to minimize RD by up to 80–97 % while improving IQ, including greater contrast-to-noise ratio (CNR) and signal-to-noise ratio (SNR) [[12], [14], [26]]. The ASIR technique with LD protocol resulted in 22–48 % reduction in RD [[10], [11], [16], [21]]. Similarly, the ADMIRE [22] and IMR[20] techniques resulted in up to a 30 % reduction in RD.

DLIR employs deep learning techniques to reduce image noise (IN), artifacts, and has been reported to decrease RD by up to 85 % while maintaining IQ [29]. Sun et al. and Lee et al. discovered that DLIR produced better IQ, particularly in lesion identification and sharpness, even at LD [[23], [29]].

Discussion

Tube voltage

The tube voltage/kilovoltage peak (kVp), plays an important role in the optimization of the radiation dose for pediatric CT head examinations. Lowering the kVp reduces the amount of scattered radiation, which contributes to the overall dose received by the patient. This reduction is particularly important in pediatric patients because of their smaller size and developing tissues. Lowering the tube voltage can result in increased IN and lead to poor IQ. However, recent advancements in IR algorithms have minimized this effect to some extent permitting reduced tube voltages without appreciably affecting the IQ. Park et al. [15] examined the effects of reduced tube-voltage (80 kVp) on IQ and radiation exposure in children undergoing CT head imaging. By using lower kVp, ATCM and IR techniques, the CTDIvol, DLP and effective dose were reduced by 4.7 %, 6.9 % and 6.5 % respectively compared with those of the standard 120 kVp protocol. Similar findings were reported by Priyanka et al., who used 80 and 100 kVp along with a low tube current and an IR technique, and the RD was reduced by 68 % and 58 % in the < 1 year and 1–5 years age groups respectively. The study also noted that there was improved gray – white matter CNR at 80 kV[30].

The automatic kVp technique enables the scanner to evaluate patient attenuation and select an optimal kVp (70–140 kVp) that maintains a balance between RD and IQ. Although this approach has been highly effective in reducing RD for pediatric body CT, its performance in pediatric head CT has been inconsistent [40], [41]. Papadakis et al. [18] reported a 70 % RD reduction in pediatric head CT angiography using automatic kVp, whereas its use in non-contrast pediatric head CT resulted in a 21 % increase in RD. When combined with an automatic tube current (mA), however, automatic kVp appears to be more beneficial. Morel et al. [13] reported a 21 % RD reduction in infant head CT examinations when a combined approach was applied.

Tube current-exposure time product (mAs)

Reducing the mAs can lower the RD, but may increase image noise, similar to lowering the tube voltage. In a study by Udayashankar et al., LD scans were performed by reducing mAs (220–––80 mAs) which resulted in a 63 % RD reduction with higher IN and lower CNR in LD images than in SD images. Thus, the study concluded that the low mAs CT head protocol can provide diagnostically acceptable IQ and can be used over SD CT in children with multiple scans in the future[7].

Few studies have reported lowered mAs on the basis of age, body mass, head circumference and indications. In a study by Priyanka et al. low mAs were used on the basis of age (< 1 year: 150 mAs, 1–5 years: 200 mAs). The study noted significant dose reduction (< 1 year: 68 %, 1–5 years: 58 %) and the IQ was maintained by using IR techniques (iDose⁴) [30]. In a study by Jonczyk-Potoczna et al., a low mAs protocol was used on the basis of body mass (< 20 kg: SD-190mAs, LD-100mAs; ≥ 20 kg: SD-350 mAs, LD- 200 mAs).The study revealed that with the use of the low mAs technique, the overall DLP was reduced by 40 % compared with that of the SD technique while the diagnostic IQ was retained [8]. Bingyang et al., used a low mAs protocol on the basis of head circumference. The study reported that a head circumference based low mAs protocol can substantially reduce the organ specific radiation dosage with diagnostically acceptable images [24]. In a study by Morton et al., LD protocol was used on the basis of indications such as a half dose (50 % reduction in mAs) to evaluate ventricles, catheter placement in patients with shunted hydrocephalus and quarter dose (75 % reduction in mAs) for postoperative craniosynostosis imaging. The other scanning parameters were kept constant for both LD protocols. IN increased and resulted in poor IQ as the tube current decreased. To reduce the IN, a postprocessing package “Neuro 3D Filter N2” (GE Healthcare) was used in both LD protocols [9].

Scan length and scan time

In CT, the RD is exactly proportional to the scan length. One effective method to minimize RD during a CT examination is to shorten the scan length [42]. On the basis of clinical indications, the scan length for a paediatric CT head examination can be tailored. The study by Eddy et al. reported that the scan length for the indication of head trauma must be limited to the second cervical spine instead of the fourth cervical spine with the exception of multiple trauma cases, which results in a significant reduction in RD [25].

Shortening the scan time also helps in optimizing the RD. It also helps reduce motion artefacts which eliminate the use of sedation in pediatric patients. However, the use of a very short scan time can result in increased image noise due to a decrease in the amount of data used for image reconstruction, necessitating the use of high kV and mAs. Most modern CT vendors provide a range for selecting the scan time, such as 0.8 s – 0.5 s. However, for CT exams, a 0.5 s scan period is used to balance radiation exposure and IQ [43], [44].

Pitch and collimation

By increasing the pitch or by using narrow collimation the scan time can be reduced which leads to a reduction in RD. However, increasing the pitch leads to poor IQ with higher noise levels and lowers the CNR and spatial resolution. Therefore, a higher pitch can be used in conjunction with the IR technique which can reduce the RD even for the LD protocol without affecting the IQ [[19], [45]].

Iterative reconstruction (IR) techniques

The common method of reducing RD involves decreasing kV and mAs. However, it causes a rise in noise and produces low-quality images which is mainly due to the drawback of FBP. This shortcoming of FBP has been responsible for the evolution of IR algorithms. RD to patients can be optimized using this IR approach without deteriorating the quality of the image for different body regions, including pediatric CT examinations [46], [47], [48], [49].

The studies by Vorona et al [10], Kim et al. [16] and Sherif et al.[21] reduced mA was used for the ASIR group for pediatric patients undergoing CT head examinations resulting in a 13 % − 42 % reduction in RD for different age groups (0–15 years). The studies also noted less IN with higher CNR and higher qualitative IQ. Similarly, McKnight et al. [11] reported that with the use of ASIR, the CTDIvol was reduced by 28 % and 48 % for 3–12 years and > 12 years age groups, respectively with no difference in IQ. Kaasalainen et al. [12], Ernst et al. [14], and Atri et al [26] assessed the usefulness of MBIR approaches combined with ultra-low-dose CT protocols (reduced kV and mAs) for imaging CT brain. The studies noted that with the use of ultra-low dose protocol with MBIR, the RD was reduced by 83––97 % compared with that of the SD protocol along with higher quantitative and qualitative IQ. Cho et al investigated the feasibility of ADMIRE technique with a LD protocol by reducing the mAs to 50, 55 and 80 for less than 1 year, 1–12 year and more than 12 years, respectively for pediatric CT head. For LD protocol, FBP images presented greater IN compared with images reconstructed different levels of ADMIRE. The average reductions in CTDIvol, DLP and ED with the LD protocol were 30.6 %, 32.1 and 32.1 % respectively, compared with the SD protocol [22]. Southard et al. analysed the RD and IQ of pediatric CT head in FBP and IMR images. They reported that qualitative and quantitative IQ was superior in IMR group than in the other groups. The authors also noted a significant reduction in reconstruction time and CTDIvol such as 100 s and 24.4 mGy and 147 s and 31.1 mGy for the IMR and FBP groups respectively [20]. Similar findings were reported by Rabinowich et al(34) who reported that LD images reconstructed by IMR had higher CNR, SNR with fewer artifacts than FBP images.

Chang et al., [17] conducted a phantom study to establish a LD scanning protocol for adult and pediatric CT head examinations. The optimal scanning protocol for pediatric CT head was 80 kVp and 50mAs for maxillary sinus, brainstem and 80 kVp and 300 mAs for temporal bone. The study concluded that the IR technique (iDose⁴ level 3) helps in setting the optimum CT head scanning protocol for adults and pediatric patients where the dose is optimized without affecting the IQ. Similar findings were reported by Muhammad et al.,[27] who studied LD phantoms (by reducing mAs), and reported a 20 % decrease in RD with increased IQ. In the study by Priyanka et al. [30], an age-based LD protocol was used for pediatric CT head examinations, images reconstructed using iDose4 level 3 demonstrated higher quantitative and qualitative IQ compared to the SD protocol.

Deep learning image reconstruction (DLIR) technique

Compared with FBP, the IR technique increases the IQ by reducing the image noise and artifacts for the LD protocol [50], [51], [52], [53]. However, for LD protocol, IR at higher levels results in an artificial or plastic-looking appearance in the image which lowers IQ and the ability to diagnose pathologies. [54], [55], [56]. These limitations of the IR technique led to the development of the DLIR technique.

Multiple studies have indicated that AI-based deep learning reconstruction (DLIR) can enhance IQ in pediatric CT while allowing substantial RD reduction compared with IR methods [[57], [58], [59]]. Evidence from both phantom and clinical studies has demonstrated a reduced IN, improved CNR, and maintenance of diagnostically acceptable IQ [[23], [29]].

Sun et al., [23] used an LD protocol at 120 kV with 120 – 176 mAs. The images were reconstructed with 5 mm and 0.625 mm slice thicknesses with FBP, 50 % ASIR-V, 100 % ASIR-V and DLIR-high. The 5 mm DLIR −high images showed lowest IN and 0.625 mm DLIR images showed similar IN (3.11 ± 0.58 HU) and overall IQ scores (3.04 ± 0.33) as 5 mm 50 % ASIR −V images (3.16 ± 0.60 HU and 3.05 ± 0.23). However, 0.625 mm DLIR- high images presented greater lesion detection (69 lesions) compared to 5 mm 50 % ASIR-V images (65 images). Hence, with the use of DLIR −high with similar image quality and slice thickness for pediatric CT head examination 85 % reduction in radiation dose can be achieved. A similar study was conducted by Lee et al., [29] in which age-based protocols such as 100 kVp, 125mAs for less than 1 year, 100kVp, 145 mAs for 1–6 years and 120 kVp and 100 – 135 mAs for 6–12 years were used. All the images were reconstructed with FBP, ASIR-V and DLIR (low, medium and high). Compared with ASIR-V, the high level DLIR significantly improved image quality with a higher SNR, CNR, grey matter – white matter differentiation, sharpness (p < 0.05) and reduced image noise. Among all the evaluated dose reduction techniques, AI-based DLIR consistently provided the greatest improvement in image quality while enabling substantial RD reduction. Compared with conventional IR and protocol adjustments, DLIR reduces image noise, enhances CNR, and maintains diagnostically acceptable quality.

This scoping review mapped the range of RD reduction strategies in pediatric CT head imaging, including IR, AI-enhanced reconstruction, and protocol optimization. Phantom studies generally report greater relative dose reductions than clinical studies, reflecting differences between controlled experiments and real-world practice [[12], [17], [18], [27]]. While most strategies have demonstrated technical feasibility, limitations such as small sample sizes, heterogeneous protocols, and inconsistent reporting have been noted, which may affect generalizability.

The clinical applicability of these strategies varies by setting. In resource-limited environments, older scanners and a lack of AI tools may restrict implementation, highlighting the need for pragmatic, low-cost protocol adaptations. In contrast, high-technology settings can leverage advanced hardware and software to optimize doses without compromising image quality. These findings emphasize tailoring dose reduction strategies to available resources and identifying key research gaps to guide safe and effective implementation across diverse clinical contexts.

Limitations

Many studies are limited by small sample sizes, lowering their generalizability across various populations, such as different age groups and clinical indications. Furthermore, there is a lack of uniformity of pediatric CT head protocols, with variations in scanner models, dose metrics, and reconstruction techniques, making comparisons across studies challenging. The trade-off between LD protocols and diagnostic IQ is another concern, as many studies have not fully analysed whether reduction techniques provide sufficient IQ for clinical conditions. In addition, the possibility of publication bias cannot be excluded, as studies with negative or inconclusive findings may be underrepresented. Non-English studies were excluded because of limitations in translation resources and to ensure accurate data extraction. Advancements such as AI based DLIR techniques are not widely available, resulting in a lack of research on their usefulness in low-resource settings. These limitations highlight the need for multicenter studies with standardized LD DLIR protocols that strike a balance between dose reduction and diagnostic efficacy.

Conclusion

As the pediatric population is more sensitive to the effects of radiation, advanced dose reduction techniques for pediatric CT head imaging are important for minimizing RD and balancing the diagnostic IQ. The IR and DLIR techniques have been proven to increase the overall IQ with reduced image noise and improved lesion detection even for LD CT head protocols. Thus, IR and DLIR have the potential to improve patient care and diagnostic accuracy in pediatric CT head imaging.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: