Machine learning in orthopaedic surgery

Table 1 Summary of machine learning for orthopaedic surgery risk assessment

Ref.	Conclusion
Bevevino et al[26]	ANN capable of accurately estimating the likelihood of amputation
Gowd et al[25]	Supervised ML outperformed ASA classification models in predicting adverse events, transfusion, extended length of stay, surgical site infection, return to operating room, and readmission
Harris et al[24]	ML was moderately accurate in predicting 30-d mortality and cardiac complications after elective primary TJA
Kim et al[23]	ANN more accurate than ASA in predicting mortality, VTE, cardiac and wound complications following posterior lumbar spine fusion

ML: Machine learning; ANN: Artificial neural network; ASA: American Society of Anesthesiologists; TJA: Total joint arthroplasty; VTE: Venous thromboembolism.

Table 2 Summary of machine learning for orthopaedic surgery outcomes assessment

Ref.	Conclusion
Bongers et al[40]	ML algorithm overestimated ability to predict 5-year survival in patients with chondrosarcoma
Fontana et al[41]	Used ML to demonstrate fair-to-good ability in predicting 2-year postsurgical MCID following TJA
Greenstein et al[51]	Used EMR-integrated ANN to predict discharge disposition after TJA on small data set
Janssen et al[38]	Boosting ML algorithm far superior in training data sets to classic scoring system and nomogram in predicting survival in patients with long bone metastases at 30 days, 90 days, and 1 year
Karnuta et al[50]	Bayes ML algorithm demonstrated excellent accuracy in prediction of length of stay and cost of an episode of care for hip fracture
Menendez et al[44]	Used ML on patient-narrative analysis to show patient satisfaction after TSA is linked to hospital environment, nontechnical skills, and delays
Navarro et al[46]	Created a valid ML algorithm that predicted length of stay and costs before primary TKA
Pereira et al[55]	Boosting ML algorithm comparable to nomogram in its ability to predict survival in metastatic spine disease with testing data sets
Ramkumar et al[45]	Created a valid and reliable ML algorithm that predicted length of stay and payment prior to primary THA
Ramkumar et al[47]	Developed several ML based models for primary LEA that preoperatively predict cost, length of stay, and discharge disposition
Thio et al[39]	Created a high performing ML algorithm that could predict 5-year survival in patients with chondrosarcoma

ML: Machine learning; MCID: Minimally clinically important difference; TJA: Total joint arthroplasty; EMR: Electronic medical record; ANN: Artificial neural network; TSA: Total shoulder arthroplasty; TKA: Total knee arthroplasty; THA: Total hip arthroplasty; LEA: Lower extremity arthroplasty.

Table 3 Summary of machine learning for orthopaedic surgery imaging applications

Ref.	Subspecialty	Conclusion
Al-Helo et al[66]	Spine	Neural network (93.2% accurate) and k-means approach (98% accurate) used on CT scans for segmentation and prediction of lumbar wedge fractures
Forsberg et al[62]	Spine	Annotated MRIs with information labels for each spine vertebrae used to accurately detect (99.8%) and label (97%) cervical and lumbar vertebrae
Hetherington et al[64]	Spine	CNN successfully identified lumbar vertebral levels on ultrasound images of the sacrum
Jamaludin et al[65]	Spine	CNN model achieved 95.6% accuracy comparable to experienced radiologists in disc detection and labeling of T2 weighted sagittal lumbar MRIs
Pesteie et al[63]	Spine	Used ML system to detect laminae and facet joints in ultrasound images to assist in epidural steroid injection and facet joint injection administration
Ashinsky et al[71]	Joints/arthritis	ML algorithm predicted clinically symptomatic OA on T2 weighted maps of central medial femoral condyle with 75% accuracy
Liu et al[72]	Joints/arthritis	CNN performed rapid and accurate cartilage and bone segmentation within the knee joint
Shah et al[73]	Joints/arthritis	CNN used to automate the segmentation and measurement of cartilage thickness based on MRIs of healthy knees
Xue et al[70]	Joints/arthritis	CNN model trained to diagnose hip OA comparable to an attending physician with 10 years of experience in diagnosing hip OA
Kruse et al[75]	Trauma	ML improved hip fracture detection beyond logistic regression using dual x-ray absorptiometry
Olczak et al[74]	Trauma	DL networks identified fracture, laterality, body part, and exam view on orthopaedic trauma radiographs of the hand, wrist, and ankle
Oh et al[78]	Oncology	ML showed superior predictive accuracy in predicting pathological femoral fractures in metastatic lung cancer

ML: Machine learning; CNN: Convolutional neural network; OA: Osteoarthritis.

Table 4 Summary of machine learning for orthopaedic surgery basic science applications

Ref.	Application	Conclusion
Begg et al[83]	Gait analysis	Used SVM to automate recognition of gait changes due to aging
Joyseeree et al[84]	Gait analysis	Used random forest, boosting, and SVM to identify disease on gait analysis data
Sikka et al[85]	Wearable technology	Utilized ML analytics via wearable technology to improve sports performance and identify risk factors for injury in sports
Cilla et al[86]	Implant design	ML techniques used to optimize short stem hip prosthesis to reduce stress shielding effects and achieve better short-stemmed implant performance

ML: Machine learning; SVM: Support vector machine.