One-click AI-assisted analysis of bright- and black-blood LGE and T2 mapping in acute STEMI patients

Presenting Author(s)

• 

  Victor de Villedon de Naide, MSc

  PhD student
  Bordeaux University - INSERM U1045
  Bordeaux, Aquitaine, France

Primary Author(s)

• 

  Victor de Villedon de Naide, MSc

  PhD student
  Bordeaux University - INSERM U1045
  Bordeaux, Aquitaine, France

Co-Author(s)

• EG

  Edouard Gerbaud, MD, PhD

  Cardiac Intensive Care Unit, Groupe Hospitalier Sud, CHU de Bordeaux, Pessac, France, France
• SV

  Sane Viola

  Engineer
  IHU Liryc, Université de Bordeaux, France
• 

  Thaïs Genisson, MSc

  PhD student
  IHU Liryc, Université de Bordeaux
  Bordeaux, Aquitaine, France
• 

  Kalvin Narceau, MSc

  Phd student
  Bordeaux University - INSERM U1045
  Pessac, Aquitaine, France
• EB

  Ewan Barel

  Engineer
  IHU Liryc, Université de Bordeaux, France
• TR

  Théo Richard, MSc

  Engineer
  IHU LIRYC, Heart rhythm disease institute, Université de Bordeaux – INSERM U1045, Avenue du Haut Lévêque, 33604, Pessac, France
  Bordeaux, Aquitaine, France
• PJ

  Pierre Jaïs, MD, PhD

  PROF/PhD
  Hôpital Cardiologique du Haut-Lévêque, CHU de Bordeaux
  Bordeaux, Aquitaine, France
• MS

  Matthias Stuber, PhD

  Professor/Director
  CIBM/CHUV/UNIL
  Lausanne, Switzerland
• HC

  Hubert Cochet, MD, PhD

  Professor
  Bordeaux University - INSERM U1045
  Bordeaux, Aquitaine, France
• 

  Aurelien Bustin, FSCMR

  Research Associate
  Department of Cardiovascular Imaging, Hôpital Cardiologique du Haut-Lévêque, CHU de Bordeaux, Avenue de Magellan, Pessac, France; IHU LIRYC, Electrophysiology and Heart Modeling Institute, Université de Bordeaux – INSERM U1045, Avenue du Haut Lévêque, Pessac, France; Department of Diagnostic and Interventional Radiology, Lausanne University Hospital and University of Lausanne, Lausanne, Switzerland
  Bordeaux, Aquitaine, France

Background: Cardiac magnetic resonance (CMR) is the reference standard for evaluating post-infarction prognostic markers after acute STEMI¹. Recent developments, such as co-registered joint bright- and black-blood LGE and T2 mapping (SPOT-MAPPING), have the potential to provide both anatomical and tissue characterization in a single acquisition². However, image analysis remains time-consuming and user-dependent. The integration of artificial intelligence (AI)-based analysis could enable rapid, reproducible, and fully automated assessment, therefore facilitating clinical adoption and allow large-scale research. This study aimed to evaluate the feasibility, accuracy, and efficiency of an AI-assisted SPOT-MAPPING pipeline in acute STEMI patients, compared to expert manual analysis.

Methods: Sequence: SPOT-MAPPING employs an ECG-triggered 2D single-shot sequence alternating between bright- and black-blood LGE images². Incremental T2-preparation durations in the bright-blood beats (0, 27, 35, 40, 55ms) enable generation of T2 maps, yielding co-registered datasets for anatomy, scar, and oedema assessment within a single acquisition (Fig.1).

Processing: Automated left ventricular (LV) segmentation and right ventricular insertion points detection were performed on bright-blood images, using combined Transformer³ and U-Net⁴ models trained on 400 independent patients⁵. LV contours were then propagated to black-blood images for automated scar segmentation (histogram-based thresholding), and to T2 maps for T2 quantification according to the AHA 16-segment model (Fig.1).

Experiments: Ten acute STEMI patients (9 male, age 58±14yo) underwent CMR at 1.5T (Siemens Magnetom Aera) 4-7 days after primary PCI. Protocol included reference T2 mapping, PSIR, and SPOT-MAPPING sequences. Agreement between automated and manual analysis was assessed using Dice coefficients, paired t-tests, Bland-Altman analysis, and bias (95% CI). Parameters including scar volume (g), burden (%LV mass) and transmurality (%), T2 values at the segmental, slice, and patient levels, processing time and segmentation failure rates were measured.

Results: High agreement with manual reference was achieved for automated LV wall segmentation (Dice 87.2±1.3%, slice failure rate 5.9%) (Table 1). Scar segmentation accuracy was moderate-to-high (Dice 68.3±8.2%, slice failure rate 30%). T2 quantification demonstrated excellent agreement at segment, slice and patient levels (mean bias: -2.5ms to -2.9ms). Automated analysis reduced processing time from >20min (manual) to < 0.1min. Representative segmentations and T2-map Bullseyes (Fig. 2) confirmed precise localisation of scar and oedema.

Conclusion: AI-assisted SPOT-MAPPING enables fast and automated quantification of myocardial scar and oedema in a single CMR acquisition in acute STEMI patients. Although scar segmentation failures remain a limitation, these findings support further refinement and large-scale validation in clinical and research applications.

Quantitative comparison of manual and automated SPOT-MAPPING for myocardial scar and oedema assessment. Comparison of LV wall volume, Dice scores, scar volume, scar burden, scar transmurality, T2 values at segment/slice/patient levels, segmentation failure rates, and processing times, with bias and 95% confidence intervals. Abbreviations: LV, left ventricle; STEMI, ST-elevation myocardial infarction; N/A, not applicable.

	Manual Segmentation	Automated Segmentation	P-value	Bias (95% CI)	Notes
LV wall volume (mL)	95.9 ± 29.3	92.1 ± 28.1	0.737	-3.8 (-5.8 to -1.8)	-
LV wall Dice score (%)	N/A	87.2 ± 1.3	N/A	N/A	vs. manual ground truth
Scar volume (mL)	17.6 ± 20.1	10.9 ± 8.5	0.285	-6.7 (-9.6 to -3.7)	-
Scar burden (% of LV mass)	31.8 ± 13.1	28.7 ± 6.6	0.510	-3.1 (-5.7 to -0.6)	-
Scar transmurality (%)	68.0 ± 8.0	63.0 ± 7.5	0.172	-5.0 (-7.7 to -2.2)	-
Scar Dice score (%)	N/A	68.3 ± 8.2	N/A	N/A	vs. manual ground truth
T2 values – segment basis (msec)	61.9 ± 12.5	59.3 ± 12.3	0.037	-2.5 (-2.8 to -2.3)	-
T2 values – slice basis (msec)	63.0 ± 11.2	60.3 ± 10.5	0.076	-2.7 (-3.1 to -2.4)	-
T2 values – patient basis (msec)	63.6 ± 6.4	60.8 ± 6.8	0.277	-2.9 (-3.5 to -2.3)	-
Failure rate scar segmentation (% affected slices)	0.0%	30.0%	N/A	N/A	defined as unusable segmentation
Failure rate LV segmentation (% affected slices)	0.0%	5.9%	N/A	N/A	defined as unusable segmentation
Processing time (min)	>20.0	< 0.1	< 0.001	N/A	-

AI-assisted SPOT-MAPPING pipeline for automated myocardial scar and oedema characterization in acute STEMI. Schematic overview of the co-registered acquisition of bright-blood, black-blood, fused, and T2 map images, and the automated analysis steps including LV segmentation, scar segmentation, and T2 quantification and corresponding 3D reconstructions showing spatial distribution. Abbreviations: AI, artificial intelligence; LV, left ventricle; STEMI, ST-elevation myocardial infarction.
Representative examples of automated versus manual quantification of myocardial scar and oedema using SPOT-MAPPING. A) Representative basal, mid-ventricular, and apical slices from SPOT-MAPPING, including bright-blood, black-blood, and T2 maps, demonstrate regional detection of myocardial oedema and scar, with 3D reconstructions showing spatial distribution. B) Comparison of automated vs. manual segmental T2 quantification in multiple patients, with bullseye plots and scatter plots showing high correlation and agreement across T2 values. C) Scar segmentation overlays from SPOT-MAPPING (manual and automated) compared to PSIR reference images in multiple patients, with Dice coefficients demonstrating strong concordance between automated and manual methods.