Time-driven Survival Analysis from FDG-PET/CT in Non-Small Cell Lung Cancer

Sambit Tarai; Ashish Chauhan; Elin Lundström; Johan Öfverstedt; Therese Sjöholm; Veronica Sanchez Rodriguez; Håkan Ahlström; Joel Kullberg

cs.CVApr 8, 2026

Time-driven Survival Analysis from FDG-PET/CT in Non-Small Cell Lung Cancer

Authors: Sambit Tarai, Ashish Chauhan, Elin Lundström, Johan Öfverstedt, Therese Sjöholm, Veronica Sanchez Rodriguez, Håkan Ahlström, Joel Kullberg

Organizations: Radiology, Department of Surgical Sciences, Uppsala University, Uppsala, SE-75185, Sweden · Molecular Imaging and Medical Physics, Department of Surgical Sciences, Uppsala University, Uppsala, SE-75185, Sweden · Antaros Medical, Mölndal, SE-43153, Sweden

Abstract

Purpose: Automated medical image-based prediction of clinical outcomes, such as overall survival (OS), has great potential in improving patient prognostics and personalized treatment planning. We developed a deep regression framework using tissue-wise FDG-PET/CT projections as input, along with a temporal input representing a scalar time horizon (in days) to predict OS in patients with Non-Small Cell Lung Cancer (NSCLC). Methods: The proposed framework employed a ResNet-50 backbone to process input images and generate corresponding image embeddings. The embeddings were then combined with temporal data to produce OS probabilities as a function of time, effectively parameterizing the predictions based on time. The overall framework was developed using the U-CAN cohort (n = 556) and evaluated by comparing with a baseline method on the test set (n = 292). The baseline utilized the ResNet-50 architecture, processing only the images as input and providing OS predictions at pre-specified intervals, such as 2- or 5-year. Results: The incorporation of temporal data with image embeddings demonstrated an advantage in predicting OS, outperforming the baseline method with an improvement in AUC of 4.3%. The proposed model using clinical + IDP features achieved strong performance, and an ensemble of imaging and clinical + IDP models achieved the best overall performance (0.788), highlighting the complementary value of multimodal inputs. The proposed method also enabled risk stratification of patients into distinct categories (high vs low risk). Heat maps from the saliency analysis highlighted tumor regions as key structures for the prediction. Conclusion: Our method provided an automated framework for predicting OS as a function of time and demonstrates the potential of combining imaging and tabular data for improved survival prediction.

Figures & tables

Figure 1: Overview of the proposed framework for automated OS prediction: [a] generation of tissue-wise PET/CT projections, [b] imaging feature extraction using a CNN, [c] temporal feature extraction using a FNN, [d] incorporation of temporal data with image embeddings, [e] classification of OS status for a given time.

Figure 2: Data sampling strategy used during training of the proposed framework. [a] For deceased patients, random time-points were sampled from two phases: the alive phase (time between scan date and the date of death) and the deceased phase (time between the date of death and LFD). [b] For alive patients, random time-points were sampled between the scan date and LFD.

Comparison criteria (for each network)	Deep regression (ResNet-50)
Comparison criteria (for each network)	baseline	proposed
[1] Number of networks	n	1
[2] Trainable parameters for each network	24,680,769	33,077,569
[3] Model size (MB) for each network	296	397
[4] Training time per epoch (seconds)	10	225
[5] Validation time per epoch (seconds)	2	126

Table 1: Comparison criteria for each network between baseline and proposed methods for OS prediction across various criteria. In both the methods ResNet-50 was used as the backbone architecture. Here "n" corresponds to number of networks used (In our case, n = 10).

Figure 3: Illustration of OS probability over time for patients in different risk categories: (a) Low-risk patient: OS probability remains high and stable (high AUSPC). (b) High-risk patient: OS probability drops quickly (low AUSPC).

Experimentation	AUC at pre-specified time intervals (in years)
	.5	1	1.5	2	2.5	3	3.5	4	4.5	5	Mean
(1) CPH (clin + IDP)	.583	.580	.647	.648	.662	.664	.657	.683	.677	.693	.649
(2) DeepHit (clin + IDP)	.750	.706	.727	.740	.736	.694	.662	.669	.641	.604	.693
(3) DeepSurv (clin + IDP)	.773	.719	.729	.734	.735	.719	.711	.718	.719	.714	.717
(4) DeepHit (imaging)	.751	.705	.701	.677	.675	.662	.662	.666	.666	.660	.683
(5) DeepSurv (imaging)	.724	.684	.698	.698	.693	.684	.676	.679	.678	.678	.689

Table 2: Comparison between the baseline and proposed methods in predicting OS status at pre-specified time intervals from the test set.

Experimentation	Inputs ( $CT^{\text{AIP}}$ / $PET^{\text{MIP}}$ / $SEG$ )					AUC	C-Index	Accuracy
	org	bone	lean	adipose	air
1 org PET	$\times$ / $\checkmark$ / $\times$	$\times$ / $\times$ /-	$\times$ / $\times$ /-	$\times$ / $\times$ /-	$\times$ / $\times$ /-	.759	.657	.670
2 org CT	$\checkmark$ / $\times$ / $\times$	$\times$ / $\times$ /-	$\times$ / $\times$ /-	$\times$ / $\times$ /-	$\times$ / $\times$ /-	.758	.613	.683
3 all PET	$\times$ / $\checkmark$ / $\times$	$\times$ / $\checkmark$ /-	$\times$ / $\checkmark$ /-	$\times$ / $\checkmark$ /-	$\times$ / $\checkmark$ /-	.770	.635	.677
4 all CT	$\checkmark$ / $\times$ / $\times$	$\checkmark$ / $\times$ /-	$\checkmark$ / $\times$ /-	$\checkmark$ / $\times$ /-	$\checkmark$ / $\times$ /-	.752	.575	.677
5 all PET/CT	$\checkmark$ / $\checkmark$ / $\times$	$\checkmark$ / $\checkmark$ /-	$\checkmark$ / $\checkmark$ /-	$\checkmark$ / $\checkmark$ /-	$\checkmark$ / $\checkmark$ /-	.796	.643	.705

Table 3: Results of ablation experiments using the proposed method on the test set, illustrating the impact of various tissue-wise FDG-PET/CT input channels on OS prediction.

Figure 4: (a) Predicted OS probabilities as a function of time for a patient where the green line represents the alive phase (GT), the red segment represents the deceased phase (GT), and the intersection of the black line with the x-axis indicates the predicted time of death. (b) Kaplan-Meier curve illustrating GT vs predicted OS probabilities on the test set.

Figure 5: Kaplan–Meier curve illustrating risk stratification of NSCLC patients in the test set, based on model-predicted survival probabilities, dividing them into high-risk (n=40) and low-risk (n=252) groups.

Figure 6: Kaplan–Meier curves illustrating risk stratification of NSCLC patients within each T-stage in the cross-validation set using the proposed method: (a) T1 (high-risk = 15; low-risk = 165), (b) T2 (high-risk = 26; low-risk = 89), (c) T3 (high-risk = 22; low-risk = 51), and (d) T4 (high-risk = 37; low-risk = 57), based on model-predicted survival probabilities.

Figure 7: Overview of saliency analysis from the proposed method: [a] PET MIP, [b] Tumor location, [c] Heatmap highlighting regions influencing the neural network’s decision.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

	Cross-val (missing)	Test (missing)
Total	556	292
Demographics
Age (years)	70.31 $\pm$ 8.62	69.81 $\pm$ 10.31
Sex	556	292
Male	239	127
Female	317	165

Appendix

Table 4: Data description for NSCLC patients in the U-CAN cohort.

Figure 8: Overview of predicted OS probabilities as a function of time on the test set for patients who were deceased ([a]-[d]) and patients who remained alive ([e]-[h]) as per LFD. Solid green line represents the patient in ground truth (GT) alive phase. Solid red lines represents the patient in GT deceased phase. Dotted green lines indicate the future forecast for patients in the alive phase. Dotted red lines indicate the future forecast for patients in the deceased phase. The point where the solid black line intersects the x-axis indicates the predicted time of death.

Figure 9: Kaplan-Meier curves illustrating overall survival (OS) probabilities as a function of time on the test set for the following comparisons: [a] low vs high TMTV based on median cut-off value, [b] female vs male.

Cancer Survival Prediction

Cancer survival prediction estimates patient outcomes and time to events from clinical, molecular, and medical-imaging data.

See all 92 papers

Time-Conditioned and Multi-Time Survival Prediction from 2D PET/CT Projections in Lung Cancer
Jun 10, 2026Ashish Chauhan, Sambit Tarai, Elin Lundström +3Cancer Survival Prediction Survival Analysis
Radiology, Department of Surgical Sciences, Uppsala University, Uppsala, Sweden · National Academic Infrastructure for Supercomputing (NAISS), LinköpingJun University, Linköping, Sweden · Antaros Medical, Mölndal, Sweden +1
Structured Proxy Features for Multimodal NSCLC Survival Prediction from Pretreatment CT
Aug 1, 2026Huu Phong Nguyen, Delower Hossain, Ehsan Saghapour +2Cancer Survival Prediction Oncology
Department of Biomedical Informatics and Data Science, School of Medicine, The University of Alabama at Birmingham (UAB), Birmingham, AL, United States. · Department of Computer Science, The University of Alabama at Birmingham (UAB), Birmingham, AL, United States. · System Pharmacology and AI Research Center (SPARC), The University of Alabama at Birmingham (UAB), Birmingham, AL, United States.
CT-CLIP Representations for Multimodal Lung Cancer Survival Prediction
Jul 9, 2026Sofie Allgöwer, Mikael Johansson, Andreas Hallqvist +4Medical Imaging Foundation Models Radiology VLMs
Chalmers University of Technology, Gothenburg, Sweden · Umeå University, Umeå, Sweden · Sahlgrenska University Hospital, Gothenburg, Sweden +1

Jun 10, 2026J/K move · Enter open · S save

Time-Conditioned and Multi-Time Survival Prediction from 2D PET/CT Projections in Lung Cancer

Ashish Chauhan, Sambit Tarai, Elin Lundström, Johan Öfverstedt+2

Radiology, Department of Surgical Sciences, Uppsala University, Uppsala, Sweden · National Academic Infrastructure for Supercomputing (NAISS), LinköpingJun University, Linköping, Sweden · Antaros Medical, Mölndal, Sweden+1

Cancer Survival Prediction Survival Analysis PET/CT Imaging Oncology

Open paper AskPDF arXiv

Accurate prediction of overall survival (OS) from positron emission tomography/computed tomography (PET/CT) can support personalized treatment and follow-up strategies in oncology. However, the impact of temporal modeling on imaging-based survival prediction remains insufficiently explored. We investigate how different temporal formulations influence survival prediction by developing two complementary approaches: Attention-guided Time-Conditioned Survival (ATCS) and Multi-Time Survival (MTS). We retrospectively analyzed pre-treatment PET/CT images from 848 patients with non-small cell lung cancer (NSCLC), including 556 for model development and 292 for held-out testing. A previously proposed Time-Conditioned Survival (TCS) model was used as a baseline. Models were trained using 5-fold cross-validation and evaluated on the test set using time-dependent area under the curve (AUC) at 6-month intervals from 0.5 to 5 years. Both ATCS and MTS outperformed the baseline TCS model, achieving mean AUCs of 0.794 and 0.793, respectively, compared to 0.767. ATCS performed better at earlier time points (0.5-3 years), whereas MTS performed better at later intervals (3.5-5 years). Combining tumor-specific and tissue-wise PET/CT features improved performance over either input alone. Finer temporal discretization improved short-term prediction, while coarser intervals provided more stable long-term estimates. These findings demonstrate that temporal modeling and input design influence PET/CT-based survival prediction. The proposed approaches enable time-specific survival estimation from pre-treatment imaging and may support improved risk stratification and clinical decision-making.

Abstract

Cancer Survival Prediction

Explore similar work

Time-Conditioned and Multi-Time Survival Prediction from 2D PET/CT Projections in Lung Cancer

Structured Proxy Features for Multimodal NSCLC Survival Prediction from Pretreatment CT

CT-CLIP Representations for Multimodal Lung Cancer Survival Prediction