Early neural signatures of cancer‐related fatigue: A longitudinal magnetic resonance imaging study of cortical micro‐ and macrostructural changes during breast cancer chemotherapy

Abstract

Background

Cancer‐related fatigue is a prevalent and debilitating symptom in patients with breast cancer (BC) undergoing chemotherapy (CTx), yet its early neural correlates remain incompletely understood.

Methods

This longitudinal study included 61 female patients with BC and 41 healthy female controls assessed before CTx and after two cycles of CTx, or at matched intervals. Assessments included 3T brain magnetic resonance imaging, patient‐reported fatigue and quality‐of‐life measures, clinician‐rated affective symptom scales, and an objective cognitive screening measure. Surface‐based analyses were used to extract cortical microstructural measures, including cortical mean diffusivity (cMD), and macrostructural measures. Linear mixed‐effects models were used to examine longitudinal changes and group‐by‐time interaction effects in clinical, patient‐reported, cognitive, and cortical structural outcomes, as well as their associations during early CTx.

Results

During early CTx, patients with BC showed worsening fatigue and worsened emotional well‐being, and changes in these outcomes were positively associated. cMD decreased over time in the left isthmus cingulate cortex, and the magnitude of this decrease was associated with worsening fatigue and worsened emotional well‐being. Mediation analysis suggested an indirect association between cMD changes in this region and fatigue through worsened emotional well‐being.

Conclusion

These findings suggest that early increases in fatigue during CTx are linked to worsening emotional well‐being and region‐specific cortical microstructural changes. The results highlight a potentially informative neural‐affective pathway in the early emergence of fatigue, while requiring replication and further validation before clinical translation.

Short abstract

Early chemotherapy in patients with breast cancer was associated with worsening fatigue and emotional well‐being, accompanied by decreased cortical mean diffusivity in the left isthmus cingulate cortex. These findings suggest a potential early psycho–neural pathway linking cortical microstructural alterations to cancer‐related fatigue through emotional well‐being.

INTRODUCTION

Cancer‐related fatigue (CRF) is the most prevalent symptom among breast cancer (BC) patients undergoing chemotherapy (CTx). ¹ Unlike ordinary tiredness, CRF is a persistent and disproportionate sense of cognitive and physical exhaustion that disrupts daily functioning, reduces quality of life, and may compromise treatment adherence. ² , ³ Although CRF often emerges during CTx and can persist long after treatment, its neural mechanisms remain insufficiently understood.

CRF is thought to arise from interacting systemic and neurobiological processes, including vagal dysregulation, circadian disruption, and proinflammatory cytokine cascades. ⁴ , ⁵ Among these mechanisms, CTx‐induced central neurotoxicity may be particularly important. Through neuronal DNA damage, cerebral microvascular injury, myelin loss, and inflammatory responses, CTx can alter brain structure and functional connectivity, thereby affecting regions involved in cognitive processing and emotional regulation. ³ , ⁶ , ⁷

Neuroimaging studies support this view, showing abnormal brain metabolism, ⁸ altered prefrontal activation, ⁹ white matter microstructural abnormalities, ¹⁰ and reduced structural integrity ¹¹ after CTx, all of which have been associated with fatigue severity. Bekele et al. ¹² further reported reduced local efficiency in brain networks after CTx that correlated negatively with fatigue scores. However, most prior studies have focused on the subacute (1 week to 6 months) or delayed (>6 months) post‐CTx periods, ¹³ whereas early acute‐phase neural alterations remain underexplored. Evidence suggests that structural and functional brain changes may occur within 1 month of CTx initiation, ¹⁴ , ¹⁵ , ¹⁶ but whether these early changes track the dynamic progression of CRF has not been systematically examined.

Previous studies of fatigue‐related neural substrates have mainly used voxel‐based or surface‐based morphometry to assess macrostructural indices, such as gray and white matter volume and cortical thickness (CTh). ¹⁷ , ¹⁸ , ¹⁹ Although informative, these measures may be less sensitive to early neural injury because overt macroscopic alterations may require longer time to become apparent after neuronal loss, glial apoptosis, or microvascular damage. ¹³ , ²⁰ By contrast, early cellular‐level changes involving membrane permeability, myelin integrity, and blood–brain barrier function may alter water diffusion before overt cortical atrophy becomes detectable. ¹⁹ , ²⁰ , ²¹ , ²² Microstructural cortical metrics may therefore complement macrostructural measures in characterizing treatment‐related brain alterations. ²³

Cortical mean diffusivity (cMD), derived from surface‐based diffusion‐weighted imaging (DWI) analysis, provides an indirect marker of cortical microstructural integrity by quantifying water diffusion within the cortex. cMD can detect microstructural abnormalities even in the absence of cortical thinning or gray matter loss and has been applied to prodromal neurodegenerative disease. ²⁴ , ²⁵ , ²⁶ , ²⁷ Increased cMD is likely to reflect freer water diffusion related to neuronal structural damage, whereas decreased cMD is generally considered to reflect restricted diffusion, potentially associated with neuroinflammatory cell infiltration. ²² In Alzheimer’s disease, cMD decreases during the early preclinical stage and subsequently increases in later preclinical and symptomatic stages, ²⁵ , ²⁸ changes linked to astrocyte activation and synaptic degeneration. ²⁹ These findings suggest that cMD may help capture early cortical remodeling related to CRF during CTx.

Based on this evidence, we hypothesized that BC patients would show early CTx‐induced cortical alterations at both macrostructural and microstructural levels, and that these alterations would be associated with cognitive and emotional changes contributing to CRF progression. Using a single‐center prospective longitudinal cohort design, we combined macrostructural cortical measures with the microstructural metric cMD to characterize early cortical changes and their relationship with CRF, and further used mediation analysis to explore whether emotional changes mediated the association between regional cortical alterations and fatigue.

MATERIALS AND METHODS

Participants

This prospective longitudinal study enrolled 68 women with newly diagnosed unilateral invasive BC from the Breast Tumor Center of Hefei Cancer Hospital, Chinese Academy of Sciences, between August 2022 and May 2025; these patients constituted the BC group. All patients underwent surgery followed by postoperative adjuvant CTx, with regimens selected in routine clinical practice according to standard indications and individual clinicopathologic characteristics, including molecular subtype, disease stage, HER2 status, and patient tolerance. No patient had received preoperative treatment or had distant metastases. Forty‐five age‐matched healthy women were recruited from the local community as the healthy control group (HC group). Eligibility criteria included female sex, right‐handedness, Chinese proficiency, and no personal history of psychiatric or neurological disorders. Exclusion criteria were structural brain abnormalities, alcohol or substance abuse, psychiatric or neurodegenerative disease, major visual or hearing impairment, poor magnetic resonance imaging (MRI) quality due to artifacts or excessive motion, and unwillingness to complete follow‐up MRI. BC patients underwent baseline MRI and clinical, patient‐reported, and cognitive assessments during the week before adjuvant CTx initiation (time point 1 [tp1]). Follow‐up assessments were completed within 3 weeks after two cycles of CTx (time point 2 [tp2]). The HC group was assessed using the same protocol at comparable intervals (Figure 1A). Demographic data were collected by questionnaire, and clinical and treatment data were obtained from electronic medical records. The study was approved by the institutional ethics committee and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

FIGURE 1.

Study design, imaging processing, and statistical workflow. (A) Timeline of scales and image acquisition. (B) T1w images were processed using FreeSurfer’s longitudinal pipeline to generate a within‐subject template and derive CTh, MC, and LGI. DWI data were preprocessed and fitted to generate MD maps, which were registered to the corresponding T1w images, projected onto the cortical surface, and sampled at the mid‐cortical depth to obtain regional cMD values. (C) Statistical analysis process. BC indicates breast cancer; cMD, cortical mean diffusivity; CTh, cortical thickness; CTx, chemotherapy; DWI, diffusion‐weighted imaging; HC, healthy controls; LGI, local gyrification index; MC, mean curvature; MD, mean diffusivity; T1w, T1‐weighted imaging; tp, time point.

Clinical, patient‐reported, and cognitive assessments

Before MRI, all participants completed a multimodal assessment battery. Fatigue was assessed using the validated Chinese version of the Brief Fatigue Inventory (BFI). The BFI score was calculated as the mean of the nine items, ranging from 0 to 10, with higher scores indicating greater fatigue. ³⁰ Based on commonly used BFI categories, scores of 1–3 were considered mild, 4–6 moderate, and 7–10 severe fatigue; scores ≥4 indicated clinically elevated fatigue. ³¹ Objective cognition was assessed using raw Montreal Cognitive Assessment (MoCA) total scores as a continuous cognitive screening measure. Depressive and anxiety symptoms were evaluated using the Hamilton Depression Rating Scale (HAMD) and Hamilton Anxiety Rating Scale (HAMA). Breast cancer–specific health‐related quality of life was assessed in BC patients only using the Functional Assessment of Cancer Therapy‐Breast (FACT‐B), which comprises physical well‐being, social/family well‐being, emotional well‐being, functional well‐being, and breast cancer–specific concerns domains. ³² In the standard FACT‐B scoring system, higher scores indicate better quality of life. For analytic consistency with the other patient‐reported symptom measures in the present study, FACT‐B item responses were recoded to derive adverse‐direction domain scores: positively worded items were reverse‐scored at the item level, whereas negatively worded symptom‐burden items were retained in their original direction. Accordingly, higher recoded domain scores indicated poorer well‐being or greater breast cancer–specific symptom burden. All assessments were administered by the same associate chief physician with more than 3 years of experience in psychological assessment.

MRI acquisition and processing

Participants were instructed to keep their eyes closed, remain still, and stay awake during MRI acquisition. Earplugs and foam padding were used to reduce scanner noise and head motion. High‐resolution T1‐weighted (T1w) images were acquired using a three‐dimensional magnetization‐prepared rapid gradient‐echo sequence: repetition time (TR)/echo time (TE) = 2000/3.4 ms, 1 mm isotropic voxel size, flip angle = 8°, field of view = 240 × 240 mm², matrix size = 256 × 256, and 170 gapless transverse slices. Diffusion‐weighted images were obtained using a 32‐direction single‐shot echo‐planar imaging protocol: TR/TE = 8496/90 ms, 2 mm isotropic voxel size, b = 1000 s/mm², and 66 contiguous slices. A single b = 0 image was also acquired with an anterior–posterior phase‐encoding direction.

T1w images were processed using the FreeSurfer longitudinal stream (version 6.0; https://surfer.nmr.mgh.harvard.edu) to derive cortical macrostructural morphology metrics, including CTh, mean curvature (MC), and local gyrification index (LGI). ³³ , ³⁴ , ³⁵ Processing included skull stripping, Talairach registration, intensity normalization, bias‐field correction, tissue segmentation, cortical surface reconstruction, and regional metric extraction (Figure 1B). The longitudinal stream generates an unbiased within‐subject template and initializes each time point from this template, reducing across‐visit variability. ²⁹ T1w images and cortical surfaces were visually inspected.

Diffusion MRI was processed using a surface‐based cortical diffusion pipeline. Raw data were denoised using MRtrix3 (version 3.0.4) and corrected for Gibbs‐ringing artifacts. ³⁶ FSL (FMRIB Software Library, version 6.0.7.3) was then used for brain extraction, correction of head motion and eddy current‐induced distortions, and diffusion tensor fitting with dtifit to generate voxel‐wise mean diffusivity (MD) maps. Because inverse phase‐encoding data were unavailable, dedicated susceptibility‐distortion correction could not be performed. To improve diffusion‐to‐anatomy correspondence, the b0 image was nonlinearly registered to the corresponding T1w image acquired in the same session using Advanced Normalization Tools (ANTs, version 2.5.1), followed by FreeSurfer boundary‐based registration (bbregister) refinement. ²⁹ The resulting transforms were used to map MD images into the individual T1w anatomical space for cortical surface sampling.

Quality control combined quantitative motion assessment and visual inspection: mean absolute motion from the FSL EDDY movement root‐mean‐square output was used as the primary motion metric; data sets with mean absolute motion ≥2 mm were excluded, ³⁷ , ³⁸ severe artifacts, extensive outlier slices, or unsatisfactory diffusion‐to‐T1 alignment were excluded. cMD maps were generated by projecting MD values onto the cortical surface using FreeSurfer’s mri_vol2surf at the mid‐cortical depth to reduce cerebrospinal fluid and white matter contamination. ³⁹ Regional cMD values were extracted from the 68 Desikan‐Killiany cortical regions, and the same longitudinally propagated atlas was used to extract CTh, MC, and LGI at each time point.

Statistics

All analyses were conducted using R (version 4.5.0). Between‐group demographic differences were assessed using independent‐samples t tests or Mann‐Whitney U tests, as appropriate, and χ² tests for categorical variables.

Linear mixed‐effects models (LMMs) were used to assess longitudinal changes and group‐by‐time interaction effects for clinical, patient‐reported, cognitive, and cortical structural outcomes (Figure 1C). Fatigue, depressive symptoms, anxiety symptoms, objective cognitive function, and cortical structural metrics served as dependent variables. Fixed effects included group (BC group vs. HC group), time (tp1 vs. tp2), the group‐by‐time interaction, and age and education as covariates. Subject ID was modeled as a random intercept to account for within‐subject variability across repeated measures. Because FACT‐B domain scores were assessed only in the BC group, these outcomes were analyzed separately using LMMs with time as a fixed effect, age and education as covariates, and subject ID as a random intercept. Baseline group differences in cortical metrics were also examined after adjustment for age and education. Multiple comparisons were controlled using the false discovery rate (FDR) method, and effect sizes are reported as Cohen’s d with 95% confidence intervals (CIs).

Partial correlation analyses controlling for age and education examined associations between fatigue and other clinical or patient‐reported outcomes showing significant longitudinal or group‐by‐time effects, and between longitudinal changes in significant cortical regions and these outcomes. Two‐sided p < .05 was considered statistically significant. Mediation analyses tested whether outcomes significantly associated with both cMD and fatigue mediated their relationship, with age and education as covariates. Path coefficients were estimated using ordinary least‐squares regression, and the indirect effect was evaluated using bias‐corrected bootstrap 95% CIs based on 1000 resamples.

RESULTS

Sample characteristics

A total of 113 participants were recruited at tp1. During follow‐up and quality control, seven patients with BC were not included in the final longitudinal analysis: three were lost to follow‐up, two discontinued because of severe gastrointestinal adverse reactions, and two were excluded because diffusion MRI head motion exceeded the prespecified threshold (Table S1). Four HC participants were also not included in the final longitudinal analysis: one was diagnosed with cancer during follow‐up and three withdrew before tp2. The final longitudinal sample included 61 patients with BC and 41 HC participants. Diffusion MRI head motion did not differ significantly between groups (Table S2).

Adjuvant CTx regimens were heterogeneous, reflecting routine clinical treatment rather than protocol assignment. Forty‐one of 61 patients received cyclophosphamide‐containing regimens. An exploratory sensitivity analysis comparing patients who received cyclophosphamide‐containing regimens with those who did not showed no evidence that the principal longitudinal cortical findings differed by regimen type (Table S3). Age and menopausal status did not differ between groups, whereas education showed a borderline between‐group imbalance (Table 1).

TABLE 1.

Demographic and clinical information.

Parameters	BC group (n = 61)	HC group (n = 41)	z / χ2	p
Age, years	52 (44–57)	53 (30–60)	–0.94	.349 a
Education, No. (%)
Primary school or below (≤6 years)	29 (47.5)	14 (34.2)	5.92	.052 b
Secondary education (7–12 years)	19 (31.2)	9 (21.9)
College or above (>12 years)	13 (21.3)	18 (43.9)
Menstruation, No. (%)
Premenopause	21 (34.4)	20 (48.8)	2.1	.147 b
Menopause	40 (65.6)	21 (51.2)
Subtype, No. (%)
Luminal A	5 (8.2)
Luminal B	16 (26.2)
HER2‐enriched	28 (45.9)
Basal‐like	12 (19.7)
AJCC stage, No. (%)
I	4 (6.6)
II	41 (67.2)
III	16 (26.2)
Chemotherapy regimen, No. (%)
EC	16 (26.2)
AC	2 (3.3)
TC	23 (37.7)
THP	6 (9.8)
TCbHP	14 (23.0)
Interval from tp1 to tp2, days	68 (61–74)	64 (60–67)	1.93	.054 a

Note: Continuous variables are presented as median (Q1–Q3); categorical variables are presented as No. (%).

Abbreviations: AC, doxorubicin (adriamycin) + cyclophosphamide; AJCC, American Joint Committee on Cancer; BC, breast cancer; EC, epirubicin + cyclophosphamide; HC, healthy controls; HER2, human epidermal growth factor receptor 2; TC, docetaxel + cyclophosphamide; TCbHP, docetaxel + carboplatin + trastuzumab + pertuzumab; THP, docetaxel + trastuzumab + pertuzumab; tp, time point.

^a Indicates Mann–Whitney U test.

^b Indicates χ² chi‐square test.

Clinical, patient‐reported, and cognitive outcomes

The proportion of BC patients meeting the prespecified threshold for clinically elevated fatigue increased from 18 of 61 (29.5%) at tp1 to 43 of 61 (70.5%) after two cycles of CTx, whereas no HC participant met this threshold at either time point. From tp1 to tp2, the BC group showed significant within‐group worsening in fatigue, depressive symptoms, and anxiety symptoms, with all p values <.001, whereas the HC group showed no significant changes. Significant group‐by‐time interactions were observed for all three outcomes, indicating different longitudinal trajectories between groups. MoCA scores showed no significant within‐group changes or group‐by‐time interaction (Table 2). Within the BC group, recoded FACT‐B physical well‐being (p < .001), emotional well‐being (p = .039), and breast cancer–specific concern scores (p = .001) increased from tp1 to tp2, indicating worsening outcomes, whereas social/family and functional well‐being scores did not change significantly (Table 3).

TABLE 2.

Fatigue, affective symptoms, and objective cognitive assessment results.

Outcome	BC group		HC group		Between‐group difference, p		Within‐group hange, p		Group‐by‐time interaction, p
	tp1	tp2	tp1	tp2	tp1	tp2	BC group	HC group	BC group vs HC group
Fatigue (BFI score)	2.4 ± 2.7	6.0 ± 3.2	0.3 ± 0.2	0.3 ± 0.2	<.001	<.001	<.001	.942	<.001
Depression (HAMD score)	7.5 ± 6.0	9.9 ± 5.8	2.7 ± 1.6	2.0 ± 1.4	<.001	<.001	.001	.428	.007
Anxiety (HAMA score)	7.4 ± 6.6	10.9 ± 6.9	2.6 ± 1.6	2.1 ± 1.5	<.001	<.001	<.001	.679	.003
Cognition (MoCA score)	25.8 ± 2.0	26.0 ± 2.0	26.4 ± 2.3	26.3 ± 2.2	.728	.662	.331	.754	.386

Note: Data are presented as means ± SD. All p values are FDR‐adjusted from age‐ and education‐adjusted linear mixed‐effects models.

Abbreviations: BC, breast cancer; BFI, Brief Fatigue Inventory; FDR, false discovery rate; HAMD, Hamilton Depression Rating Scale; HAMA, Hamilton Anxiety Rating Scale; HC, healthy controls; MoCA, Montreal Cognitive Assessment; tp, time point.

TABLE 3.

Breast cancer–specific patient‐reported quality‐of‐life outcomes (FACT‐B) in the BC group.

FACT‐B domains	tp1	tp2	t	p value
Physical well‐being	7.3 ± 3.4	10.6 ± 4.7	5.69	<.001
Social/family well‐being	20.7 ± 4.1	21.5 ± 3.7	1.87	.193
Emotional well‐being	7.2 ± 3.1	8.5 ± 3.7	2.32	.039
Functional well‐being	14.5 ± 6.0	16.1 ± 5.6	2.16	.105
Breast cancer–specific concerns	7.8 ± 3.7	9.6 ± 4.2	3.73	.001

Note: Data are means ± SD. Higher FACT‐B scores indicated poorer well‐being or greater symptom burden after reverse‐scoring positively worded items. All p values are FDR‐adjusted from age‐ and education‐adjusted linear mixed‐effects models.

Abbreviations: BC, breast cancer; FACT‐B, Functional Assessment of Cancer Therapy‐Breast; FDR, false discovery rate; tp, time point.

Cortical structural changes

At baseline, no significant between‐group differences in cMD, CTh, MC, or LGI were observed after adjustment for age and education (Table S4).

For cMD, the BC group showed significant within‐group longitudinal decreases in the left isthmus cingulate cortex (L‐IsthC), lateral occipital cortex (L‐LOC), parahippocampal gyrus (L‐PHG), pars opercularis (L‐ParsOp), and temporal pole (L‐TP), as well as the right cuneus (R‐CUN), pars opercularis (R‐ParsOp), posterior cingulate cortex (R‐PCC), and temporal pole (R‐TP). Significant within‐group longitudinal increases were observed in the left and right caudal middle frontal gyri (L/R‐cMFG) and the right banks of the superior temporal sulcus (R‐BSTS). No significant within‐group cMD changes were observed in the HC group. Significant group‐by‐time interactions were detected for L‐IsthC, L‐PHG, L‐TP, R‐ParsOp, and R‐PCC, indicating greater cMD decreases in the BC group than in the HC group (Figure 2A; Tables 4 and 5).

FIGURE 2.

Cortical structural changes. (A) Significant within‐BC longitudinal changes in cMD, CTh, MC, and LGI. Blue and red indicate decreases and increases, respectively. (B) Significant group‐by‐time interactions showing regions with different trajectories between the BC and HC groups. Blue and red indicate greater decreases and increases in the BC group, respectively. (C) Overlay of within‐BC changes and interaction effects. Light blue indicates within‐BC changes only; orange indicates overlapping effects. BC indicates breast cancer; cMD, cortical mean diffusivity; CTh, cortical thickness; HC, healthy controls; LGI, local gyrification index; MC, mean curvature.

TABLE 4.

Results of longitudinal changes in cortical metrics within the BC group.

Comparison	Brain region	t	p	Cohen's d (95% CI)
cMD
tp1 > tp2	L‐IsthC	–2.59	.019	–0.52 (–0.67 to –0.37)
	L‐LOC	–2.93	.017	–0.58 (–0.73 to –0.43)
	L‐PHG	–2.91	.017	–0.57 (–0.73 to –0.42)
	L‐ParsOp	–2.85	.017	–0.57 (–0.72 to –0.42)
	L‐TP	–2.32	.024	–0.56 (–0.71 to –0.41)
	R‐CUN	–2.40	.022	–0.31 (–0.46 to –0.17)
	R‐ParsOp	–2.01	.048	–0.26 (–0.40 to –0.12)
	R‐PCC	–2.74	.018	–0.53 (–0.68 to –0.38)
	R‐TP	–2.42	.022	–0.31 (–0.45 to –0.17)
tp1 < tp2	L‐cMFG	2.61	.019	0.52 (0.37–0.67)
	R‐cMFG	2.67	.018	0.56 (0.41–0.71)
	R‐BSTS	2.04	.046	0.40 (0.26–0.55)
CTh
tp1 > tp2	L‐MTG	–3.11	.010	–0.40 (–0.54 to –0.25)
	L‐STG	–2.58	.013	–0.33 (–0.47 to –0.19)
	L‐SMG	–2.30	.032	–0.29 (–0.44 to –0.15)
	R‐cACC	–3.32	.009	–0.72 (−0.88 to −0.57)
	R‐BSTS	–2.58	.021	–0.33 (–0.48 to –0.19)
MC
tp1 < tp2	R‐cACC	2.46	.030	0.50 (0.35–0.65)
	R‐CUN	2.75	.023	0.61 (0.46–0.77)
	R‐EC	2.18	.039	0.28 (0.14–0.43)
LGI
tp1 < tp2	L‐BSTS	2.54	.017	0.57 (0.42–0.72)
	R‐CUN	2.31	.035	0.55 (0.40–0.70)
	R‐EC	2.02	.047	0.52 (0.37–0.67)
	R‐FG	2.14	.036	0.44 (0.29–0.58)

Note: The table presents within‐group longitudinal changes in cortical metrics in the BC group. All p values were FDR‐corrected for multiple comparisons. For cMD, tp1 > tp2 indicates declining cortical mean diffusivity over time.

Abbreviations: BC, breast cancer; BSTS, banks of the superior temporal sulcus; cACC, caudal anterior cingulate cortex; CI, confidence interval; cMD, cortical mean diffusivity; cMFG, caudal middle frontal gyrus; CTh, cortical thickness; CUN, cuneus; EC, entorhinal cortex; FDR, false discovery rate; FG, fusiform gyrus; IsthC, isthmus cingulate cortex; L, left; LGI, local gyrification index; LOC, lateral occipital cortex; MC, mean curvature; MTG, middle temporal gyrus; ParsOp, pars opercularis; PCC, posterior cingulate cortex; PHG, parahippocampal gyrus; R, right; SMG, supramarginal gyrus; STG, superior temporal gyrus; TP, temporal pole; tp, time point.

TABLE 5.

Results of group‐by‐time interaction effects on cortical metrics.

Comparison	Brain region	t	p	Cohen's d (95% CI)
cMD
	L‐IsthC	–2.61	.025	–0.53 (–0.68 to –0.38)
	L‐PHG	–2.81	.022	–0.57 (–0.72 to –0.42)
	L‐TP	–2.23	.042	–0.45 (–0.60 to –0.30)
	R‐ParsOp	–2.10	.047	–0.43 (–0.57 to –0.28)
	R‐PCC	–2.74	.022	–0.55 (–0.70 to –0.40)
CTh
	L‐MTG	–2.77	.027	–0.56 (–0.71 to –0.41)
	L‐STG	–2.16	.033	–0.44 (–0.58 to –0.29)
	L‐SMG	–2.44	.032	–0.49 (–0.64 to –0.34)
	R‐cACC	–2.24	.034	–0.45 (–0.60 to –0.30)
MC
	R‐cACC	2.96	.015	0.54 (0.39–0.69)

Note: Results show group‐by‐time interaction effects on cortical metrics from linear mixed‐effects models adjusted for age and education. All p values were FDR‐corrected for multiple comparisons.

Abbreviations: BC, breast cancer; cACC, caudal anterior cingulate cortex; CI, confidence interval; cMD, cortical mean diffusivity; CTh, cortical thickness; FDR, false discovery rate; HC, healthy controls; IsthC, isthmus cingulate cortex; L, left; MC, mean curvature; MTG, middle temporal gyrus; ParsOp, pars opercularis; PCC, posterior cingulate cortex; PHG, parahippocampal gyrus; R, right; SMG, supramarginal gyrus; STG, superior temporal gyrus; TP, temporal pole.

For macrostructural cortical metrics, the BC group showed significant longitudinal CTh decreases in the left middle temporal gyrus (L‐MTG), left superior temporal gyrus (L‐STG), left supramarginal gyrus (L‐SMG), right caudal anterior cingulate cortex (R‐cACC), and R‐BSTS. Significant group‐by‐time interactions for CTh were observed in L‐MTG, L‐STG, L‐SMG, and R‐cACC. MC increased longitudinally in the BC group in R‐cACC, R‐CUN, and the right entorhinal cortex (R‐EC), with a significant group‐by‐time interaction only in R‐cACC. LGI increased longitudinally in the BC group in the left banks of the superior temporal sulcus (L‐BSTS), R‐CUN, R‐EC, and right fusiform gyrus (R‐FG), but no LGI measure showed a significant group‐by‐time interaction after FDR correction. No significant within‐group longitudinal changes in CTh, MC, or LGI were detected in the HC group (Figure 2B,C; Tables 4 and 5).

Correlation and mediation analyses

After adjustment for age and education, fatigue changes in the BC group were positively associated with emotional well‐being changes (r = 0.44, p = .002), but not with depressive symptoms, anxiety symptoms, physical well‐being, or breast cancer–specific concerns (Figure 3A). Decreased L‐IsthC cMD was associated with greater fatigue (r = –0.41, p = .005) and worse emotional well‐being (r = –0.34, p = .036) after FDR correction (Figure 3B‐C). Worsening emotional well‐being significantly mediated the association between decreased L‐IsthC cMD and increased fatigue (indirect effect: β = –0.12; 95% CI, –0.29 to –0.04, p = .033), accounting for 28.1% of the total effect (Figure 4A,B).

FIGURE 3.

Partial correlations among cMD, fatigue, and emotional well‐being in the BC group. (A) Fatigue worsening was associated with worsening emotional well‐being. (B) Greater decreases in L‐IsthC cMD were associated with greater fatigue worsening. (C) Greater decreases in L‐IsthC cMD were associated with worsening emotional well‐being. Partial correlations were adjusted for age and education. Displayed p values are FDR‐corrected. Higher emotional well‐being scores indicate worse well‐being. *p < .05; **p < .01. BC indicates breast cancer; cMD, cortical mean diffusivity; FDR, false discovery rate; L‐IsthC, left isthmus cingulate cortex; tp, time point.

FIGURE 4.

Mediation analysis. (A) Conceptual one‐mediator model showing the total effect (c), direct effect (c′), and indirect effect (a × b). (B) Mediation model in the BC group, with L‐IsthC cMD change as the independent variable, emotional well‐being change as the mediator, and fatigue change as the outcome. BC indicates breast cancer; CI, confidence interval; cMD, cortical mean diffusivity; L‐IsthC, left isthmus cingulate cortex.

DISCUSSION

In this longitudinal study, we examined early cortical microstructural and macrostructural changes in patients with BC during the initial phase of adjuvant CTx and their associations with CRF progression. The results showed longitudinal decreases in cMD across multiple cortical regions in patients with BC, accompanied by macrostructural changes including reduced CTh. Among these regions, the L‐IsthC demonstrated a greater decline over time in BC patients than in the HC group and showed moderate associations with fatigue and emotional well‐being. Mediation analysis further suggested that microstructural changes in the L‐IsthC were indirectly related to fatigue through worsened emotional well‐being. Together, these findings provide preliminary neuroimaging evidence that the early chemotherapy period is accompanied by cortical microstructural alterations linked to the emergence of fatigue‐related symptoms.

This study found that BC patients exhibited reductions in cMD across multiple cortical regions during the early CTx, suggesting that cortical microstructural alterations may emerge within the first two treatment cycles rather than only after treatment completion. Decreased cMD typically reflects restricted water diffusion, which may be associated with early neuroinflammatory responses such as cellular swelling, altered membrane permeability, or inflammatory cell infiltration. ⁴⁰ Such interpretations are biologically plausible in the context of CTx‐induced inflammatory responses, microvascular perturbation, and astrocyte activation. ⁴¹ In contrast to previous studies that primarily identified cortical abnormalities after CTx completion or in long‐term survivors, ⁴² these findings extend prior work by suggesting that cortical microstructural alterations may be detectable during the early CTx period. Accordingly, cMD should be interpreted as a sensitive but indirect DWI‐derived proxy of cortical microstructural alterations. Although susceptible to residual partial‐volume effects, cMD may provide complementary information about early tissue‐level changes beyond conventional T1‐based macrostructural measures, offering a novel perspective on early CTx‐related cortical changes.

In addition to cMD alterations, cortical morphometric changes were observed, including reduced CTh and altered MC or LGI in selected regions. Their limited spatial overlap with cMD findings suggests that early treatment‐related cortical effects may differ across imaging markers and assessment intervals. ⁴³ Prior studies indicate that cMD may capture early cortical microstructural alterations that precede or accompany macrostructural changes, ²⁶ , ²⁷ , ⁴⁴ consistent with longitudinal BC MRI studies reporting early CTh and gyrification changes after CTx. ¹⁵ However, given the short follow‐up after only two CTx cycles, these morphometric findings should be interpreted as early MRI‐detectable changes rather than evidence of rapid or irreversible cortical remodeling. Despite longitudinal processing and FDR correction, residual measurement variability and false‐positives cannot be excluded; therefore, macrostructural findings remain preliminary and complementary to cMD results.

Worsening fatigue during early CTx was associated with poorer patient‐reported emotional well‐being, but not with clinician‐rated depressive or anxiety symptoms, suggesting that subjective emotional well‐being may capture affective experiences accompanying fatigue beyond HAMD or HAMA scores. ⁴⁵ Notably, fatigue was already significantly elevated before CTx initiation, consistent with prior evidence that fatigue can be present before chemotherapy or even before cancer treatment begins. ⁴⁶ Together, these findings suggest that early cancer‐related fatigue may not solely represent a somatic readout of bodily burden, but may also involve central‐affective and neuroimmune processes related to stress, emotional appraisal, and subjective experience. ⁴⁷ However, because this study was not designed to distinguish somatic from neurogenic mechanisms directly, this interpretation remains tentative.

Further analysis showed that decreases in L‐IsthC cMD were significantly associated with increased fatigue severity, suggesting the involvement of this region in fatigue‐related processes. This finding is consistent with Dehsarvi et al., ⁴⁸ who reported an association between fatigue and structural connectivity linking the L‐IsthC with the left paracentral lobule. Decreases in L‐IsthC cMD were also associated with worsening emotional well‐being. Prior studies have implicated cingulate and isthmus cingulate alterations in affective and trauma‐related conditions, including cortical thinning in major depression and in women with borderline personality disorder and childhood trauma. ⁴⁹ , ⁵⁰ Given that the literature is not fully consistent in direction, laterality, or symptom‐level associations, our findings suggest that the L‐IsthC may represent a relevant cortical substrate linking emotional well‐being with fatigue progression during early CTx.

Building on these findings, the mediation analysis suggested a possible indirect association among L‐IsthC cMD changes, worsened emotional well‐being, and increasing fatigue. Specifically, decreases in cMD in the L‐IsthC were associated with worsened emotional well‐being, which in turn was associated with greater fatigue, accounting for approximately 28.1% of the total effect. This finding is consistent with psycho‐oncology evidence suggesting that poorer emotional processing and coping are associated with greater cancer‐related symptom burden, including fatigue, in breast cancer survivors. ⁵¹ From a network perspective, the L‐IsthC is anatomically related to posterior cingulate/default mode network (DMN) systems, which have been implicated in self‐referential processing, interoceptive awareness, and affective regulation. ⁵² , ⁵³ Prior studies have also linked aberrant DMN connectivity in cancer survivors to CRF, ⁵⁴ , ⁵⁵ and changes in posterior cingulate perfusion or metabolite levels after CTx to fatigue and memory decline. ⁵⁶ , ⁵⁷ However, this result should be interpreted as preliminary and hypothesis‐generating rather than as evidence that the L‐IsthC is a specific or exclusive substrate of CRF. Given the left‐lateralized and highly localized nature of this finding, the heterogeneous CTx regimens, and the observational design, the association between L‐IsthC cMD changes, worsened emotional well‐being, and fatigue requires independent replication in larger cohorts with more homogeneous treatment exposure, additional follow‐up time points, and multimodal imaging validation.

Several limitations should be noted. First, the modest sample size after attrition limited statistical power and precluded stratified analyses by CTx regimen. Because adjuvant CTx was selected according to routine clinical indications rather than assigned by protocol, treatment regimens were heterogeneous; therefore, the findings should be interpreted as longitudinal associations during the early CTx period rather than regimen‐specific effects. Second, causal inference is limited, as fatigue, emotional well‐being, and brain changes may also be influenced by diagnosis, surgery, stress, inflammation, endocrine factors, and other treatment‐related exposures. Third, the localized L‐IsthC finding and early macrostructural changes require independent replication. Finally, although mitigation procedures were applied, cMD remains susceptible to partial‐volume effects, residual susceptibility distortion, and diffusion‐to‐anatomy misregistration; therefore, these findings should be interpreted cautiously.

This prospective longitudinal multimodal MRI study provides evidence that BC patients exhibit microstructural alterations across multiple regions and region‐specific macrostructural changes within the first two cycles of CTx. These early imaging changes occurred alongside worsening CRF and emotional well‐being. In particular, decreases in L‐IsthC cMD were associated with greater fatigue and poorer emotional well‐being, and mediation analysis suggested a possible indirect association with fatigue through worsened emotional well‐being. These findings highlight L‐IsthC microstructural alteration as a candidate neural correlate of the emotion‐fatigue linkage during early CTx. Future research is needed to clarify the biological mechanisms underlying these early structural changes, establish their temporal and causal relationships with symptom evolution, and determine their potential relevance for monitoring and management of CTx‐related fatigue.

AUTHOR CONTRIBUTIONS

Yanfei Zhou: Conceptualization; methodology; formal analysis; writing—original draft; writing—review and editing. Jin Liu: Conceptualization; methodology; formal analysis; writing—original draft; writing—review and editing. Xiao Fu: Conceptualization; methodology; formal analysis; writing—original draft; writing—review and editing. Jian Fan: Data curation; writing—review and editing. Zong Cao: Data curation; writing—review and editing; software. Jing Li: Data curation; writing—review and editing. Ruiping Ye: Data curation; writing—review and editing. Hongzhi Wang: Conceptualization; methodology; writing—review and editing. Lizhuang Yang: Conceptualization; methodology; investigation; formal analysis; writing—review and editing; funding acquisition; supervision. Hai Li: Conceptualization; methodology; investigation; formal analysis; writing—review and editing; funding acquisition; supervision. All authors were involved in revising the manuscript critically for important intellectual content. All authors provided final approval of the manuscript and agree to be accountable for all aspects of the work.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supplementary Material

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.