Detecting Brain Cancer Progression Through Signatures of Wearable Accelerometry Data

Rishit Das; Barbora Barancikova; Matt Williams

cs.LGOct 27, 2026

Detecting Brain Cancer Progression Through Signatures of Wearable Accelerometry Data

Authors: Rishit Das, Barbora Barancikova, Matt Williams

Organizations: Department of Computing, Imperial College London · Department of Surgery & Cancer, Imperial College London

Abstract

Glioblastoma and other high-grade gliomas (HGG) have a poor prognosis and are associated with progressive functional decline. The standard of care to assess progressive disease is contrast-enhanced MRI, which is expensive, claustrophobic, and performed only once every 2-3 months. Wearable accelerometers offer a non-invasive alternative for continuously monitoring patient function between clinic visits. However, accelerometry data streams are typically noisy and highly patient-specific, making conventional time series methods and summary statistics poorly suited to detecting clinically meaningful change. This work aims to classify stable versus progressive disease states by applying rough path signatures for feature extraction from wrist-worn accelerometer data. We convert accelerometer recordings from HGG patients enrolled in the BrainWear study into walking-path segments and use signature kernel maximum mean discrepancy (SigMMD) to quantify change in patients' walking behaviour. We overcome the challenge of inter-patient heterogeneity by adjusting for the natural variation in a patient's walking during stable tumour periods. This patient-specific normalisation substantially improves separation, yielding a candidate biomarker of tumour progression. Our results provide proof-of-concept evidence that signature-based analysis of wearable accelerometry can distinguish stable from progressive disease states, motivating larger longitudinal studies of continuous wearable monitoring as a digital biomarker in neuro-oncology.

Figures & tables

Figure 1: Daily SigMMD values for a selected patient over two clinic windows. The shaded region shows the baseline period and the dashed line is the clinic visit date. The progressive window shows higher SigMMD values compared to the stable window.

Panel of Figure 2 — Figure 2: Activity label distributions across signature coefficients: (a) $S_{1}(x)^{2}$ , (b) $S_{2}(x)^{1,2}$ , (c) $S_{2}(x)^{2,1}$ , (d) $S_{2}(x)^{2,2}$ , with 1,000 samples per label. Activities have distinct signature coefficient distributions. Specifically, Walking (in blue) shows a different distribution than other activities.

Panel of Figure 3 — Figure 3: Distributions of raw mean SigMMD values and personalised SigMMD ratios for stable and progressive-disease visits. Raw mean SigMMD values show substantial overlap between the two groups, whereas patient-specific normalisation reveals a tendency toward higher personalised SigMMD ratios among the progressive visits.

Panel of Figure 4 — Figure 4: Performance evaluation of the personalised SigMMD classifier.

Model	AUC	Sensitivity	Specificity	Precision
Ours (Personalised Step Count - Ablation)	0.4424 (0.1871-0.7620)	0.3636 (0.1111-0.6250)	0.8000 (0.6471-0.9375)	0.5714 (0.2000-0.8889)
60 Day Step Count Benchmark	0.4242 (0.1905-0.6727)	0.0909 (0.0000-0.2857)	0.8667 (0.6922-1.0000)	0.3333 (0.0000-1.0000)
Ours (Personalised SigMMD)	0.8545 (0.6600-0.9925)	0.7272 (0.3636-1.0000)	0.8000 (0.6471-1.0000)	0.7272 (0.4000-1.0000)

Table 1: LOPO classification performance with 95% confidence intervals.

Figure 5: Comparison of AUC (with 95% confidence intervals) for the personalised SigMMD model against the step count benchmark and ablation model. Both step count approaches perform below a random classifier (AUC = 0.5, dashed line), with confidence intervals that cross it, whereas the personalised SigMMD model achieves a substantially higher AUC.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 6 — Figure 6: Distributions of raw mean SigMMD values and personalised SigMMD ratios for stable and progressive-disease visits, for the Triaxial representation.

Panel of Figure 7 — Figure 7: Performance evaluation of the triaxial representation using an ROC curve and confusion matrix.

Model	AUC	Sensitivity	Specificity	Precision
Personalised SigMMD	0.7455 (0.5125-0.9546)	0.7272 (0.400-1.000)	0.7333 (0.6111-0.9167)	0.6667 (0.4000-0.8750)

Appendix

Table 2: Classification Performance generated by Leave One Patient Out Classification for Triaxial representation.

Comparison	Signature Coefficient	ks stat	p-value	Significant
walking vs mixed	$S_{1}(x)^{2}$	0.064	0.033253	True
	$S_{2}(x)^{1,2}$	0.117	2.21E-06	True
	$S_{2}(x)^{2,1}$	0.106	2.60E-05	True
	$S_{2}(x)^{2,2}$	0.108	1.69E-05	True
walking vs sleep	$S_{1}(x)^{2}$	0.223	3.41E-22	True
	$S_{2}(x)^{1,2}$	0.148	5.73E-10	True

Appendix

Table 3: KS Test Results for ENMO representation

Comparison	Signature Coefficient	ks stat	p-value	Significant
walking vs mixed	$S_{1}(x)^{2}$	0.06	0.054627	False
	$S_{1}(x)^{3}$	0.055	0.097103	False
	$S_{1}(x)^{4}$	0.031	0.722825	False
	$S_{2}(x)^{1,2}$	0.086	0.001221	True
	$S_{2}(x)^{1,3}$	0.091	0.000503	True
	$S_{2}(x)^{1,4}$	0.084	0.001716	True

Appendix

Table 4: KS Test Results for Triaxial representation

Wearable Health Monitoring

Wearable health monitoring uses body-worn sensors and AI to track physiological and behavioral signals for continuous, personalized health assessment.

See all 111 papers

STRIDE: Spatial-Temporal Representation for Interval-conditioned Disease Evolution in Longitudinal Glioblastoma MRI
Oct 2, 2026Wenhao Guo, Changchang Yin, Pierre Giglio +3Longitudinal Medical Image Analysis Medical Image Representation Learning
Department of Computer Science and Engineering, The Ohio State University, Columbus, OH, USA · Department of Biomedical Informatics, The Ohio State University, Columbus, OH, USA · Department of Neurology, The Ohio State University, Columbus, OH, USA
Predictive Radiomics for Evaluation of Cancer Immune SignaturE in Glioblastoma: the PRECISE-GBM study
May 11, 2026Prajwal Ghimire, Junjie Li, Liu Yaou +2Clinical Prediction Oncology
School of Biomedical Engineering & Imaging Sciences, King’s College London, UK · Department of Neurosurgery, King’s College Hospital, London, UK · Department of Neuroradiology, Beijing Tiantan Hospital, Beijing, China +1
Frailty Estimation in Elderly Oncology Patients Using Multimodal Wearable Data and Multi-Instance Learning
Apr 8, 2026Ioannis Kyprakis, Vasileios Skaramagkas, Georgia Karanasiou +10Clinical Prediction Clinical Risk Prediction
ECE Dept. HMU FORTH-ICS Heraklion, Greece · FORTH-BRI Ioannina, Greece · Univ. of Ioannina (Med.) Physiology Lab Ioannina, Greece +6

Oct 2, 2026J/K move · Enter open · S save

STRIDE: Spatial-Temporal Representation for Interval-conditioned Disease Evolution in Longitudinal Glioblastoma MRI

Wenhao Guo, Changchang Yin, Pierre Giglio, Weidan Cao+2

Department of Computer Science and Engineering, The Ohio State University, Columbus, OH, USA · Department of Biomedical Informatics, The Ohio State University, Columbus, OH, USA · Department of Neurology, The Ohio State University, Columbus, OH, USA

Longitudinal Medical Image Analysis Medical Image Representation Learning Brain Tumor Classification

Open paper AskPDF arXiv

Glioblastoma (GBM), an aggressive primary brain tumor, is routinely monitored with longitudinal MRI after treatment. Distinguishing stable disease (SD), pseudoprogression (PsP), and true progression (TP) remains challenging because these states can show overlapping MRI appearances despite different temporal trajectories. Existing longitudinal methods still face challenges in modeling scan-specific spatial variability, variable follow-up intervals, and complementary information from the observed follow-up state and its longitudinal change. We propose STRIDE, a framework for spatial-temporal representation of interval-conditioned disease evolution that takes paired post-treatment MRI scans and their inter-scan interval as input and predicts SD, PsP, or TP. The lesion-prior-guided spatial representation combines SoftGate and an Adaptive-window Hierarchical Transformer (AWHT) to emphasize lesion-related regions while preserving surrounding context. The time-conditioned latent transition uses pair-level context and the actual inter-scan interval to estimate interval-dependent representation changes between visits. The observed--transition fusion integrates the transition-estimated follow-up representation with the directly observed follow-up representation to jointly characterize the follow-up state and its longitudinal change. BraTS2024 is used to develop and evaluate the lesion-prior generator, while longitudinal pretraining on LUMIERE supports transfer before downstream adaptation to Burdenko. On the Burdenko three-class task, STRIDE achieves a macro ROC--AUC of 0.816 and a macro F1-score of 0.796. These results support its potential for more reliable longitudinal post-treatment GBM state assessment.

Abstract

Wearable Health Monitoring

Explore similar work

STRIDE: Spatial-Temporal Representation for Interval-conditioned Disease Evolution in Longitudinal Glioblastoma MRI

Predictive Radiomics for Evaluation of Cancer Immune SignaturE in Glioblastoma: the PRECISE-GBM study

Frailty Estimation in Elderly Oncology Patients Using Multimodal Wearable Data and Multi-Instance Learning