AI-Driven Analysis of Longitudinal MRI Marrow Patterns Predicts Survival in Multiple Myeloma Treatment

Background/Objectives: Multiple myeloma (MM) remains incurable, with most patients eventually experiencing relapse/progression despite advances in treatment. This preliminary study evaluated the prognostic value of longitudinal changes in vertebral magnetic resonance imaging (MRI) patterns using a two-stage machine learning (2S-ML) framework. Methods: The study included 62 patients with paired spinal MRI scans acquired before and after autologous stem cell transplantation (ASCT). First, a pre-trained ResNet-18 was fine-tuned to classify vertebrae into four marrow infiltration patterns. Second, 12 handcrafted temporal descriptors were extracted to characterize longitudinal changes in MRI patterns. Ridge-penalized Cox proportional hazards regression was employed as the primary model to predict time to progression (TTP) and overall survival (OS). A random forest classifier was retained as a secondary analysis for fixed-horizon outcome prediction, and model performance was evaluated using patient-level leave-one-case-out cross-validation (LOCO-CV). Results: The 4MRI-P model achieved an average AUC of 0.8905 ± 0.0054 across 1208 vertebral MRI annotations. In the primary Cox analysis, adding longitudinal MRI features to clinical variables modestly improved the C-index for TTP (0.6950 to 0.7301) and OS (0.6801 to 0.6924). The secondary random forest analysis achieved time-dependent AUCs of 0.8291 ± 0.0844 for TTP at 36 months and 0.7396 ± 0.0776 for OS at 60 months. Kaplan–Meier analyses showed significant separation between risk groups for both outcomes (both p < 0.001). Conclusions: These preliminary findings suggest that AI-driven analysis of longitudinal vertebral MRI patterns may provide prognostic value for survival risk stratification in MM.

Authors

Institutions

Publication Details

Journal
Journal of Clinical Medicine
Published
2026-11-13
DOI
https://doi.org/10.3390/jcm15228906
Primary Topic
Multiple Myeloma Research and Treatments
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

AI-Driven Analysis of Longitudinal MRI Marrow Patterns Predicts Survival in Multiple Myeloma Treatment

Yifan Wang, Matthias Wilms, Lubomir M. Hadjiiski, Chuan Zhou et al.
3 links
Journal of Clinical Medicine
Multiple Myeloma Research and Treatments
article

AI-Driven Analysis of Longitudinal MRI Marrow Patterns Predicts Survival in Multiple Myeloma Treatment

Yifan Wang, Matthias Wilms, Lubomir M. Hadjiiski, Chuan Zhou, Qian Dong, Tao Chen
article en

Abstract

Background/Objectives: Multiple myeloma (MM) remains incurable, with most patients eventually experiencing relapse/progression despite advances in treatment. This preliminary study evaluated the prognostic value of longitudinal changes in vertebral magnetic resonance imaging (MRI) patterns using a two-stage machine learning (2S-ML) framework. Methods: The study included 62 patients with paired spinal MRI scans acquired before and after autologous stem cell transplantation (ASCT). First, a pre-trained ResNet-18 was fine-tuned to classify vertebrae into four marrow infiltration patterns. Second, 12 handcrafted temporal descriptors were extracted to characterize longitudinal changes in MRI patterns. Ridge-penalized Cox proportional hazards regression was employed as the primary model to predict time to progression (TTP) and overall survival (OS). A random forest classifier was retained as a secondary analysis for fixed-horizon outcome prediction, and model performance was evaluated using patient-level leave-one-case-out cross-validation (LOCO-CV). Results: The 4MRI-P model achieved an average AUC of 0.8905 ± 0.0054 across 1208 vertebral MRI annotations. In the primary Cox analysis, adding longitudinal MRI features to clinical variables modestly improved the C-index for TTP (0.6950 to 0.7301) and OS (0.6801 to 0.6924). The secondary random forest analysis achieved time-dependent AUCs of 0.8291 ± 0.0844 for TTP at 36 months and 0.7396 ± 0.0776 for OS at 60 months. Kaplan–Meier analyses showed significant separation between risk groups for both outcomes (both p < 0.001). Conclusions: These preliminary findings suggest that AI-driven analysis of longitudinal vertebral MRI patterns may provide prognostic value for survival risk stratification in MM.

Journal of Clinical MedicineVol. 15(22)
University of Michigan (US), Michigan Medicine (US)
National Institutes of Health, National Cancer Institute
Good health and well-being
Multiple Myeloma Research and Treatments
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon