Introduction

A cancer diagnosis imposes profound biopsychosocial challenges, particularly for adolescents and early young adults (AeYA, aged 15–25 years), a developmental phase marked by neurobiological maturation, identity formation, and heightened vulnerability to stress [1, 2]. While advancements in oncology have enabled prolonged remission for many AeYA patients [3], emerging evidence underscores a distinct subgroup with elevated vulnerability to post-remission depression [4,5,6]. These individuals exhibit subtle yet detectable neurobiological alterations that are linked to the interaction between critical developmental processes and the neurotoxic effects of cancer treatments [6,7,8]. The neurobiological changes may appear before clinical symptom onset, offering a critical window for early intervention [4].

Current depression screening tools, reliant on subjective psychosocial assessments, lack precision in identifying high-risk subgroups [9]. Neuroimaging advances reveal that metabolic dysregulation in stress- and emotion-processing regions—such as the prefrontal cortex, amygdala, and hippocampus—may serve as stable biomarkers of latent depression risk [10,11,12,13]. Notably, 18F-FDG PET/CT, a routine tool for cancer surveillance, provides an opportunity to quantify these metabolic signatures during remission [14]. By focusing on brain regions implicated in depression pathophysiology, PET/CT transcends transient psychosocial confounders, capturing intrinsic biological vulnerability that persists irrespective of future environmental triggers [15].

This study pioneers a precision psychiatry approach by targeting AeYA survivors with neurobiological susceptibility to depression. We hypothesize that early metabolic changes in key brain regions, detectable via PET/CT during remission, predominantly predict depression risk, even in the absence of overt psychosocial stressors. By integrating longitudinal neuroimaging with a calibrated radiomic nomogram, we aim to establish an objective, imaging-based screening tool to identify high-risk individuals before symptom onset. This paradigm shift - from reactive symptom management to proactive risk stratification - addresses a critical gap in survivorship care, offering a biologically grounded strategy to mitigate mid-to-long-term mental health disparities in this vulnerable population.

Methods

Participants recruitment

We performed multicenter two-cohort research to investigate early changes of brain metabolic values in 18-F FDG PET/CT in AeYA cancer survivors of malignancy, diagnosed as cancer in remission phase by attending doctors, aged 15 to 25 years old. Remission was defined as treatment completion within the past 2 years and in ongoing follow-up without further anti-cancer treatment. Participants should provide consent to receive regular follow-up within the next 3 years and have no diagnosis of depression (relied on both BDI <14 and psychiatrist evaluation) during the remission phase. Patients who received ongoing treatment for a stable disease state and patients unable to provide informed consent were excluded from the study.

Participants were enrolled in the discovery cohort at Sun Yat-Sen Cancer Center from February 2020 to March 2021, and enrolled in the validation cohort from three other tertiary hospitals from December 2020 to September 2021 in Second Affiliated Hospital of Shantou University Medical College, Cancer Hospital of Shantou University Medical College, and First Affiliated Hospital of Zhengzhou University. The human participant study followed Institutional Review Boards at the Second Affiliated Hospital of Shantou University Medical College via approved protocols and required informed consent from all participants of the study. All sampling procedures were performed according to the Helsinki Declaration. Information related to demographic variables and treatment history of participants were obtained from electronic medical records at each research setting in which all participant data were de-identified. Cohort Reporting adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist for cohort studies.

Assessments

Patients who consented to participate were followed up from the time of remission until they had an episode of depression or the end of the 36-month follow-up period. Experienced clinical interviewers, who were blind to the brain PET scan data, carried out semi-constructive interviews with the participants every 4 months since cancer remission. For baseline comparison, prior brain imaging information at the time of cancer diagnosis (historical PET brain imaging) and at the remission phase (PET brain imaging at remission), as well as demographic data were collected retrospectively on medical records and by means of interviews.

Socio-economic status was determined by household income (yearly), categorized according to the Population Census, adjusted for age, of the National Bureau of Statistics, as they have been shown to influence psychological research results in previous studies. Psychiatric history of depression or PTSD was defined as having such a history before remission. We recorded depressive levels and patient well-being at the remission stage, in which depressive levels were assessed with the Beck Depression Inventory (BDI) [16], and patient well-being was assessed with the Functional Assessment of Cancer Treatment-General (FACT-G) [17]. Both have been used extensively in depression assessment and show satisfactory reliability.

PET brain imaging

Brain 18F-FDG PET/CT scan was performed with Biograph Combined PET/CT scanners for all centers included in the study (Siemens, Erlangen, Germany). Included participants fasted for 6 h before scanning, and glucose levels were acquired before 370 MBq 18F-FDG administration with glucometers. Participants with capillary hyperglycemia ( >8 mmol/L) were banned from scanning. Before PET scanning, a free-breathing, low-dose CT scanning was performed for the best orientation of brain anatomy data and subsequent attenuation correction of combined images. To ensure internal calibration maximization in the discovery cohort, The PET images were reconstructed with the same standard ordered-subset algorithm in the discovery cohort in the historical and remission PET scans.

The brain was segmented by entropy adaptive thresholding in CT slices to outline brain rims, which were then superimposed on PET images. The pixel information of the PET data was extracted to calculate the mean standard uptake value (SUVmean) of each slice. Since many cerebral regional metabolic changes have been studied in depression and anxiety research, four regions of interest (ROI) that have been extensively evidenced to be hypometabolic in PET/CT brain research were applied as potential candidates to build the nomogram for predicting depression (parameters analyzed with e-soft, version 4.0; Siemens Medical Solutions, Hoffman Estates, IL, USA), including the ventral prefrontal cortex (VPC), hippocampus (HIPP), amygdala (AMG), and whole-brain (WB) area. Technicians of nuclear medicine outline the four ROIs of both hemispheres in the median copy of all CT slices that encompass these ROIs. The ROIs were superimposed on PET images to calculate SUVmean estimates for the four regions.

Analytic protocols

Initially, demographic data, historical, and remission brain imaging estimates were compared in order to find significant variables. Longitudinal imaging estimates were compared as before-and-after matched variables to generate metabolic changes in ROIs studied. This step evaluates early changes in regional glucose metabolism in participants with and without future incident depression. Second, significant imaging parameters enter multivariate survival analysis to test their association with the incidence of depression during the 3-year follow-up. Significant variables both in matched analysis and in survival analysis thus serve as candidates for nomogram development predicting depression incidence in the discovery cohort. Finally, external validation was performed to test predicting the ability of future onset of depression by early changes of imaging parameters in the validation cohort.

Statistics

Independent t-test was applied for comparison between the incident depression group and control group, and paired t-test was applied as a matched comparison in longitudinal studies of both groups. The chi-square test was used in the analysis of categorical data. Survival curves were estimated with the Kaplan-Meier method, with differences calculated with the log-rank test. A univariate Cox proportional hazards model was used to determine risk variables. Variables significant in univariate analysis were subject to the Cox proportional hazard model in multivariate analysis to identify independent variables.

A nomogram was formulated by using the package of rms in R version 2.14.1 (http://www.r-project.org/). The performance of the nomogram was measured by concordance index (C-index) and assessed by comparing nomogram-predicted versus observed Kaplan-Meier estimates of disease-free survival (DFS) probability. During the external validation of the nomogram, the total points of each patient in the validation cohort were calculated according to the established nomogram, then Cox regression in this cohort was performed using the total points as a factor, and finally, the C-index and calibration curve were derived based on the regression analysis.

Results

Baseline demographics

We recruited 992 AeYA patients at the remission phase in the discovery cohort, in whom 930 participants had evaluable 18-F FDG PET/CT scanning data both at the treatment-naive phase and at the remission phase to be included in the 36-month follow-up schedule. Seven patients refused to participate in the study due to personal reasons and were thus excluded. As such, at the end of the study, data were available for 923, including 440 males and 483 females, with a mean age of 20.30 ± 2.74, and the mean time of follow-up was 29.58 ± 10.01 months (100 patients who were lost to follow-up were labeled with the time of the last follow-up and analyzed as censored data in subsequent survival analysis). The cancer types were mainly lymphoma and leukemia, with a small proportion of sarcoma (37 patients), with demographic data shown in Table 1. A total of 267 participants had incident depression at the end of follow-up, making up an incidence rate of 28.91%. Patients with incident depression obtained a BDI score of 4.51 ± 2.78 at the remission stage, below the diagnostic threshold for depression ( ≥14) [18], but significantly higher than non-depressed patients’ 3.62 ± 2.27 (P < 0.001). The mean time of depression onset from remission was 16.69 ± 9.4 months.

Table 1 Demographics and psychosocial data in discovery cohort at baseline and during follow-up.
Full size table

Early changes in regional metabolic levels related to depression incidence

We first did a nested case-control comparison in brain metabolism of patients with and without depression at the remission phase, when patients ended all anti-cancer therapies. We found that all 4 regions of the brain metabolism, including WB, AMG, HIPP, and VPC, had significantly decreased levels of 18-F FDG SUVmean in patients with incident depression (all P < 0.001) (Fig. 1A). We then made a self-comparison of brain metabolism at baseline and at the remission phase in both groups of patients. We found that there was a decrease in SUV mean of AMG, HIPP, VPC, and WB from baseline to remission phase in patients with incident depression (all P < 0.001). For patients without depression, we did not observe significant change of SUVmean from baseline to remission phase. These findings support that changes in SUVmean were significantly different between patients with and without depression.

Fig. 1: Decreased levels of 18-F FDG SUVmean for WB, AMG, HIPP, and VPC metabolisms were independently correlated with incident depression.
figure 1

A Changes of brain regional metabolic values from baseline to remission in patients of discovery cohort with future onset of depression and control, shown as mean values ± 95% confidence interval. B Univariate analysis of depression onset risk associated with clinic-pathological variables and brain regional metabolic values in Cox proportional hazard model. C Multivariate analysis of depression onset risk associated with significant variables found in univariate analysis in the Cox proportional hazard model.

Full size image

Considering demographics that may confound such differences, and that major negative events happening during follow-up may foster incident depression, we then incorporate all such data as covariates into Cox proportional models to comprehensively investigate whether regional metabolic levels were predominantly associated with incident depression. In univariate tests, baseline values of brain metabolism did not correlate to incident depression, whereas remission-phase values bear a strong correlation together with other demographic variables (Fig. 1B). These factors were put into multivariate survival analysis and we found that SUVmean values of these brain regions at the remission phase were associated predominantly with outcomes during follow-up, in which lower metabolic values were associated with a higher risk of depression incidence during the 36-month follow-up period (Fig. 1C). When divided by quantiles of these SUVmean values, depression-free survival curves demonstrated significant stratification. These preliminary findings demonstrated that lower brain metabolism, even years before depression onset, was significantly associated with future risk of depression (Fig. 2).

Fig. 2
figure 2

Kaplan-Meier survival curve of depression onset associated with quartiles (1st to 4th quartile) of SUVmean values of each brain region, plotted as mean ± 95% CI (shaded areas), with the difference of depression risk calculated with the log-rank test.

Full size image

Independent validation and nomogram development

To test the generalization ability of these imaging parameters in predicting the future onset of depression, we did the independent analysis in the validation cohort by repeating the analyzing methods adopted in the discovery cohort. We finally recruited 518 patients with informed consent to participate in the 3-year follow-up study (Table 2). The mean follow-up time was 27.60 ± 10.72 months (98 patients were lost to follow-up and were analyzed as censored data). There were 176 patients with incident depression during follow-up (incidence rate of 33.92%), with a mean time of depression onset from remission of 16.4 ± 10.06 months. The self-comparison of brain metabolism at baseline and the remission phase also indicated a decrease of SUV mean of AMG, HIPP, VPC, and WB from baseline to remission phase in patients with incident depression (all P < 0.001) (Fig. 3). After incorporating all data as covariates into the Cox proportional hazards model, we found that SUVmeans of these brain regions at the remission phase were predominantly associated with the incidence of depression during follow-up, corroborating with findings in the discovery cohort.

Fig. 3
figure 3

Changes of brain regional metabolic values from baseline to remission in patients of validation cohort with future onset of depression and control, shown as mean values ± 95% confidence interval.

Full size image
Table 2 Demographics and psychosocial data in validation cohort at baseline and during follow-up.
Full size table

Since there was a significant association in both cohorts between regional SUVmean and depression incidence during follow-up, we then constructed a nomogram to visualize quantifiable contribution of each brain region SUVmean in predicting the depression-free survival (DFS) rates (6-month, 12-month, 24-month, and 36-month rates) in the discovery cohort during follow-up (Fig. 4A), which demonstrated a high agreement between observed and predicted failure risk with a C-index of 0.91. The projected value, when added together, of SUVmean onto the scale indicated the risk of DFS. The validation results of the current nomogram showed a C-index of 0.88, indicating sound accuracy of prediction. The calibration curve (Fig. 4B) showing the nomogram-predicted versus actual rate of depression in the validation cohort indicated a sound match.

Fig. 4: The Nomogram for the prediction of 3-year incident depression by the four regions of brain metabolisms was established and validated.
figure 4

The Nomogram established from the discovery cohort (A) and calibration curve plotted by validation cohort (B) for predicting the 3-year depression disease-free survival (DFS) rate using SUVmean values of brain regions found significant in survival analysis.

Full size image

Discussion

In this longitudinal, multicenter cohort study of AeYA cancer survivors, we mainly identified that specific regions of the brain undergo early metabolic changes evident on brain 18-F FDG PET/CT even years before depression symptom onset. Considering that sequential 18-F FDG PET/CT is a feasible tool for monitoring brain metabolism during cancer treatment monitoring, we tested and validated the efficacy of adopting the tool in depression screening. We found that metabolic values of several susceptible brain regions at the time of cancer remission could be soundly integrated into a calculating nomogram for onset prediction independent of baseline or follow-up risk factors. Given the high incidence of depression during the early phase of cancer survivorship in AeYA cancer survivors [3, 19], with an incident depression of 28.91 to 33.92% in our study, our finding indicated an idea of using non-invasive, clinically accessible prediction tool for early-phase depression screening.

Our findings in AeYA survivors should be contextualized within broader cancer-related depression mechanisms. While depression prevalence ranges between 10–25% across all cancer populations [20], younger age and advanced disease constitute established risk factors [21]. The elevated depression risk in our AeYA cohort aligns with reports of heightened vulnerability among survivors facing developmental disruptions during critical life transitions [19]. Notably, the predominance of hematological malignancies in this cohort (75–96% lymphoma/leukemia) merits consideration. Current evidence suggests these specific cancer types and their associated intensive chemotherapeutic regimens exert minimal confounding influence on long-term depression risk trajectories [19], consistent with our multivariate analyses showing non-significant effects of cancer type/treatment intensity on metabolic vulnerability or depression outcomes. This supports the interpretation that the observed brain metabolic signatures reflect age-specific neurobiological susceptibility rather than treatment artifacts.

Depression of AeYA cancer survivors may have a different etiology from those in the general public or adult cancer survivors [22]. Despite the immature brain structure and incomplete executive function, such patients underwent long-term physical morbidity, chemotherapy, and cancer-related neuropsychiatric decline during cancer management [3, 22, 23]. These factors may be inducible to a specific category of brain metabolic signatures. Although prior research has generally corroborated whole-brain hypometabolism in depressive episodes in co-morbid patients, how metabolic values were associated with future depression in AeYA patients remained poorly understood [10]. Notably, depression development following a cancer diagnosis or remission could be multifactorial, yet our study demonstrated that parameter changes in these regions were robust enough to be irrespective of inciting factors such as cancer relapse. Also, one key element of our research focused on the association between early changes of brain metabolism before depression onset, as prior research mainly focused on metabolic value changes of multiple brain regions during the onset of depression in other populations [10, 24]. Our findings thus suggested that a pre-depressive state of brain metabolism may exist in the remission phase of AeYA patients.

In patients with depression, PET imaging often reveals systemic dysfunction, manifested by reduced cerebral blood flow and metabolism [25, 26]. 18-F FDG PET can be used to study depression because of its ability to measure local brain glucose metabolism [27], and it has shown unique utility in patients with depression, showing particularly low glucose metabolism in cortical, subcortical, and cerebellar regions, as well as in the frontal and limbic systems [28, 29]. In tumor patients, PET showed a slowing of overall metabolism, possibly due to glucose competition between the tumor and brain tissue [30]. However, in our study, we observed that baseline metabolism, defined as treatment-naive phase imaging parameters, was generally similar between patients with and patients without future depression, suggesting that remission-phase changes of brain metabolism could arise from cancer-related reasons and indicate the development of depression.

In previous studies, the correlation analysis between abnormal brain metabolism and depression and anxiety showed that the frontal and temporal lobes were the most important brain regions, especially the frontal lobe [10, 31]. Studies have shown that decreased prefrontal glucose uptake is associated with aggression and impulsivity [32], and these behaviors are prevalent in cancer patients [33]. As one of the most important components of the limbic system, the hippocampus and amygdala are related to various functions of emotion, behavior, and long-term memory, and their reduced metabolism supports the hypothesis that limbic cortical pathway obstruction is related to the occurrence of depression [10, 34]. However, we believe that remission-phase hypometabolism in more brain regions outside of our study should be further investigated with high-resolution imaging tools due to the limited brain areas selected in our imaging protocols and the relatively low resolution of 18-F FDG PET/CT.

This study bears several limitations. Depression development is multifactorial. Although our study developed a model for screening future risks of depression in AeYA patients, several unaccounted factors in the 3-year follow-up may cause depression. For such reason, this study was limited to screening out susceptible candidates in real practice. Also, although the sample size was relatively large to derive statistical significance, it remains low for developing a robust model and as such the accuracy may not be generalizable to real practice.

Conclusions

This longitudinal, multicenter, nested case-control study identified that specific regions of the brain undergo early metabolic changes evident on brain 18-F FDG PET/CT even years before depression symptom onset. A well-validated nomogram for future depression onset prediction in AeYA cancer survivors, independent of baseline or follow-up risk factors, was established by metabolic changes of several susceptible brain regions detectable on 18-F FDG PET/CT at the time of cancer remission. Our findings indicate the potential of using this non-invasive, clinically accessible prediction tool as an adjunctive biomarker for early-phase depression risk stratification.