Research project · In progress · 2024 — 2027
AI-driven simulation for ACL surgery outcomes.
Machine learning meets in silico biomechanics to predict anterior cruciate ligament reconstruction outcomes — and to reshape how orthopaedic surgeons plan, perform, and follow up on ACL surgery.
Scroll
01 — About
Predicting what happens after surgery, before it happens.
Abstract
ACLART intersects anterior cruciate ligament reconstruction (ACLR) with AI machine learning and in silico simulation to propose a predictive model for ACLR clinical outcomes.
By integrating each patient's clinical and surgery data in machine-learning algorithms and computational simulation, the project delivers insights to improve patient selection, surgical decision-making, and pre- and post-operative management — enhancing the efficiency of ACLR procedures.
The inclusion of AI-driven in silico simulation into ACLR management is an innovative element of the design approach: finite-element results — including those of published studies — are integrated and driven by machine-learning models, correlating the outputs of AI via clinical data and via biomechanical in silico factors.
02 — Objectives
Six threads, one predictive tool.
01
Improve patient outcomes and long-term prognosis
Subjective failure risk estimation at a patient-specific level during pre-operative outcome discussions.
02
Refine surgical technique and personalize rehabilitation
AI-generated insights tailored to graft selection, tunnel orientation, and fixation devices for each patient.
03
Reduce complications and failures
Identify risk factors of ACLR procedures before the surgical team makes the final decision.
04
Build a platform for continuous research
Large-scale data analysis and modelling with AI — designed to outlive the funded phase of the project.
05
Deepen biomechanical understanding of the knee
AI-driven simulations that expose interactions between graft, bone, and boundary conditions.
06
Advance knowledge in orthopaedic surgery
Disseminated through peer-reviewed publications, conferences, and a national seminar organised with SPOT.
03 — Research plan
A five-stage pipeline from patient data to clinical decision.
Stage 01
Clinical data collection
Pre- and post-operative records, surgical reports, follow-up assessments and MRI imaging from multiple healthcare institutions and Scandinavian registers.
Stage 02
Pre-processing & data mining
Outlier removal, normalization, one-hot encoding and feature selection to extract patterns that influence ACLR outcomes.
Stage 03
Machine-learning modelling
Logistic regression, SVM, random forests, neural networks and MLP — cross-validated against k-fold and leave-one-out baselines.
Stage 04
In silico FEA integration
Finite-element analysis of graft, bone, and fixation devices validated against in vitro experimental knee replicas at INESC TEC and UA.
Stage 05
Validation & dissemination
Independent datasets, external validation, a point-of-care clinical calculator, and publication across peer-reviewed journals and conferences.
Research questions
Q1
How effectively can machine-learning algorithms predict post-operative clinical outcomes of ACLR based on pre-operative patient data?
Q2
What are the most significant predictors of successful ACLR outcomes identified through AI-driven analysis?
Q3
How does the performance of AI-driven prediction models compare to traditional regression-based methods?
Q4
Can ML algorithms accurately stratify patients into risk categories for complications after ACL surgery?
Q5
How do different surgical techniques or graft choices impact the predictive accuracy of AI-driven in silico models?
Q6
Can AI-driven in silico analysis help personalize rehabilitation protocols to optimize patient recovery?
Q7
How generalizable are these prediction models across different patient populations and surgical practices?
Q8
What are the potential limitations or biases inherent in AI-driven predictions for ACLR outcomes, and how can they be mitigated?
04 — Tasks
Five tasks. Four teams. One tool.
T1
Project management · Dissemination
Management, communication & knowledge transfer
Stakeholder coordination, an SPOT-organised national seminar, and a research protocol for orthopaedic surgeons to contribute ACLR data to the machine-learning models.
Targets
- Healthcare professionals & orthopaedic surgeons
- Researchers, industry partners and patients
- Knee & Biomechanics Sections of SPOT
Deliverables
- Project website + social channels
- National seminar and workshops
- Research collaboration protocol
T2
Clinical · Database · Statistics
ACLR surgery data collection & statistical analysis
Clinical data from Centro Hospitalar de Vila Nova de Gaia-Espinho, Hospital São Francisco–Porto, HUC, plus Swedish, Norwegian and Danish orthopaedic registers — over 1000 records from the surgical team alone.
Variables captured
- Patient-specific: gender, age, BMI, dominance
- Pre-op: ACL severity, TALS, MRI, KOOS QOL / Sports
- Surgical: graft choice, tunnel orientation, fixation
- Post-op: TALS, LKSS, IKDC-SKF
Analyses
- Comparative across graft types and techniques
- Regression of predictors vs. surgical success
- Survival for graft failure and revision rates
- Sensitivity for robustness and confounders
T3
AI · Machine learning · Clinical data
Machine learning applied to clinical-surgery data
KOOS QOL as the primary regression target, with TALS, LKSS and IKDC-SKF evaluated via a diverse regression ensemble. End goal: an in-clinic risk calculator for point-of-care outcome discussion.
Algorithms evaluated
- k-Nearest Neighbours, Decision Tree, SVR
- Gaussian Process, Kernel / Ridge / Lasso Lars
- SGD, Bayesian Ridge, Multi-Task Elastic Net
- Multi-Layer Perceptron Regression
Metrics
- Mean squared error (MSE)
- Mean absolute error (MAE)
- R² score and root MSE
- Grid search + k-fold cross-validation
T4
Biomechanics · In vitro + In silico
Machine learning applied to in silico ACLR simulation
An experimental knee replica with synthetic composite bones and ligament replicas, paired with finite-element models that must replicate experimental behaviour with high fidelity.
Material laws tested
- Compression–tension non-linearity
- Spring, elastic, hyperelastic
- Porohyperelastic
- Fibril-reinforced porohyperelastic
Variables studied
- Graft type, size, and orientation
- Tunnel placement and boundary conditions
- Fixation devices (tibia / femur)
- Gait patterns, joint and muscle loads
T5
Clinical outcomes · Revision probability
Clinical ACLR outcome predictions
Correlation of clinical and biomechanical variables — graft selection, tunnel placement, initial tension, fixation — to predict revision probability and stratify patients into risk categories for personalized planning.
Predictive outputs
- Patient-specific revision probability
- Risk-category stratification
- Minimal clinically important difference
Clinical integration
- Decision-support for surgical planning
- Continuous model monitoring and retraining
- Rehabilitation protocol adaptation
“AI-driven in silico analysis estimates subjective failure risk at a patient-specific level when discussing outcome expectations preoperatively.” From the ACLART research programme
05 — Team
Thirteen researchers across four teams.
Principal Investigator
José António de Oliveira Simões
F81E-8F49-9353
Surgical medical
José Carlos Noronha
8419-D869-FAF5
Surgical medical
Fernando Manuel Pereira da Fonseca
3C10-DDBF-97AD
Surgical medical
Orlando José de Almeida Branco Simões
2618-2B91-AA09
Surgical medical
Orlando José Reis Frazão
FF16-7804-383D
Surgical medical
Susana Cristina Ribeiro Novais
EB1B-D0B4-4A20
Surgical medical
Ricardo Jorge Teixeira de Sousa
CC1D-816B-FF31
AI / Machine learning
João Miguel Pinto Pereira da Silva
8A11-186A-3F2F
In silico biomodelling
António Manuel Amaral Monteiro Ramos
AF12-6860-6E65
In silico biomodelling
João Pedro Moreira de Oliveira
4415-3067-D56A
In vitro biomodelling
José Luís Santos
9D12-BC3E-ECAA
In vitro biomodelling
Paulo Roriz
8416-5174-26DE
Doctoral researcher
Gonçalo Duarte Nunes
B81F-9A45-6B22
06 — Outputs
Publications and dissemination.
International journal
Article (2025). Artificial intelligence-driven in silico simulation for prediction of anterior cruciate ligament reconstruction clinical outcomes. Computational and Artificial Intelligence for Mechanics and Biomechanics, 1, 45–51.
Article (2025). Insights from a computational model of the Anterior Cruciate Ligament Reconstruction. Results in Engineering, 27, 106125. doi.org/10.1016/j.rineng.2025.106215
National journal
In press (2026, Feb 25–27). Modelo experimental para medição das forças dos ligamentos na articulação do joelho intacta — modelo I. Mecânica Experimental.
International conference
Presented (2025, Sep 18–19). Artificial intelligence-driven in silico simulation for prediction of anterior cruciate ligament reconstruction clinical outcomes. 1st Conference on Computational and Artificial Intelligence for Mechanics and Biomechanics, Porto, Portugal.
Accepted (2026, Sep 29 – Oct 1). Large language models as assistive tools for outcome analysis after anterior cruciate ligament reconstruction: A proof-of-concept study. 46th SICOT Orthopaedic World Congress, Kyoto, Japan.
National conference
Accepted (2026, Feb 25–27). Modelo experimental para medição das forças dos ligamentos na articulação do joelho intacta — modelo I. 13º Congresso Nacional em Mecânica Experimental, Porto, Portugal.
07 — Partners
Coordination and research partners.
Coordination
esad.idea
esadidea.pt
Research partner
INESC TEC
inesctec.pt
Research partner
Univ. Aveiro
ua.pt
Funding
Funding programme
PITD — Co-financed by ERDF
Source fund
European Regional Development Fund
Specific objective
Develop and strengthen research and innovation capacities and the adoption of advanced technologies.
Operation type
Scientific R&D
Operation code
COMPETE2023-FEDER-00867200
Project number
15022