Artificial intelligence methods are developing rapidly in the medical field.However,the effectiveness of model training relies heavily on the support of sufficient sample sizes.Due to various constraints such as privacy,security,ethics,and costs in the medical field,it is rather difficult to obtain a large number of labeled training samples.Problems like the scarcity of rare disease cases,the lack of biological data for drug molecule mining,and the shortage of high-quality annotations for medical images significantly reduce the ability of models to learn from observed data,which in turn leads to poor prediction performance.In this context,constructing efficient learning artificial intelligence models for small sample data is of far-reaching significance both theoretically and practically.On the one hand,it can help to explore potential patterns when samples are insufficient in the early stage of new research.On the other hand,high-quality models can effectively reduce the cost of manual annotation,shorten the research cycle,and provide opportunities for solving challenging problems in medical research where it is difficult to collect a sufficient number of samples.Driven by both the expected advantages and actual needs,the research on artificial intelligence for small sample data has gradually become a highly anticipated and important research direction.This review systematically collates and summarizes the principles,advantages,disadvantages,applicable scenarios,and principal challenges associated with six artificial intelligence methods currently employed in the context of small-sample medical data,namely generative adversarial networks,graph neural networks,transfer learning,reinforcement learning,and Meta-learning.Furthermore,the review provides an extensive outlook and in-depth contemplation on the future trajectory of artificial intelligence methodologies in the realm of small sample data in medicine.

BACKGROUND:In recent years,artificial intelligence has been increasingly integrated into the diagnosis and treatment of lumbar disc herniation,enhancing the accuracy and efficiency of diagnostic procedures and diversifying therapeutic approaches.This integration has positioned artificial intelligence as a burgeoning focal point within the field.OBJECTIVE:To provide a comprehensive overview of the current applications of artificial intelligence in the diagnosis and treatment of lumbar disc herniation,to analyze the limitations of the relevant technologies.METHODS:A systematic computer-assisted literature search was performed in PubMed,CNKI,WanFang Database,and VIP Database for relevant publications regarding the application of artificial intelligence in the diagnosis and treatment of lumbar disc herniation from database inception up to August 2024.The search keywords included"lumbar disc herniation,artificial intelligence,machine learning,deep learning,big data,robot,neural network,model,algorithm."A total of 101 articles were selected based on predefined inclusion criteria and were reviewed.RESULTS AND CONCLUSION:Different artificial intelligence technologies have propelled the intelligent treatment of lumbar disc herniation,showing great potential for future development.Deep learning technology,based on advanced algorithms,constructs corresponding learning models to optimize the processing of X-ray,CT,and MRI images,achieving precise localization,identification,and analysis of degenerated intervertebral discs,and improving the accuracy of automated diagnosis.Big data technology utilizes data platforms to analyze vast medical data,simulate disease development trends,and provide a new perspective for disease assessment and prediction.Surgical robots,combined with robotic arms,3D high-definition vision systems,and 5G communication technology,support remote precise surgical operations,demonstrating significant technological advantages.In the future,the diagnosis and treatment of lumbar disc herniation by artificial intelligence will evolve towards standardization,efficiency,and precision through continuous optimization of algorithms and the professional development of data platforms.

Artificial intelligence methods are developing rapidly in the medical field.However,the effectiveness of model training relies heavily on the support of sufficient sample sizes.Due to various constraints such as privacy,security,ethics,and costs in the medical field,it is rather difficult to obtain a large number of labeled training samples.Problems like the scarcity of rare disease cases,the lack of biological data for drug molecule mining,and the shortage of high-quality annotations for medical images significantly reduce the ability of models to learn from observed data,which in turn leads to poor prediction performance.In this context,constructing efficient learning artificial intelligence models for small sample data is of far-reaching significance both theoretically and practically.On the one hand,it can help to explore potential patterns when samples are insufficient in the early stage of new research.On the other hand,high-quality models can effectively reduce the cost of manual annotation,shorten the research cycle,and provide opportunities for solving challenging problems in medical research where it is difficult to collect a sufficient number of samples.Driven by both the expected advantages and actual needs,the research on artificial intelligence for small sample data has gradually become a highly anticipated and important research direction.This review systematically collates and summarizes the principles,advantages,disadvantages,applicable scenarios,and principal challenges associated with six artificial intelligence methods currently employed in the context of small-sample medical data,namely generative adversarial networks,graph neural networks,transfer learning,reinforcement learning,and Meta-learning.Furthermore,the review provides an extensive outlook and in-depth contemplation on the future trajectory of artificial intelligence methodologies in the realm of small sample data in medicine.

BACKGROUND:In recent years,artificial intelligence has been increasingly integrated into the diagnosis and treatment of lumbar disc herniation,enhancing the accuracy and efficiency of diagnostic procedures and diversifying therapeutic approaches.This integration has positioned artificial intelligence as a burgeoning focal point within the field.OBJECTIVE:To provide a comprehensive overview of the current applications of artificial intelligence in the diagnosis and treatment of lumbar disc herniation,to analyze the limitations of the relevant technologies.METHODS:A systematic computer-assisted literature search was performed in PubMed,CNKI,WanFang Database,and VIP Database for relevant publications regarding the application of artificial intelligence in the diagnosis and treatment of lumbar disc herniation from database inception up to August 2024.The search keywords included"lumbar disc herniation,artificial intelligence,machine learning,deep learning,big data,robot,neural network,model,algorithm."A total of 101 articles were selected based on predefined inclusion criteria and were reviewed.RESULTS AND CONCLUSION:Different artificial intelligence technologies have propelled the intelligent treatment of lumbar disc herniation,showing great potential for future development.Deep learning technology,based on advanced algorithms,constructs corresponding learning models to optimize the processing of X-ray,CT,and MRI images,achieving precise localization,identification,and analysis of degenerated intervertebral discs,and improving the accuracy of automated diagnosis.Big data technology utilizes data platforms to analyze vast medical data,simulate disease development trends,and provide a new perspective for disease assessment and prediction.Surgical robots,combined with robotic arms,3D high-definition vision systems,and 5G communication technology,support remote precise surgical operations,demonstrating significant technological advantages.In the future,the diagnosis and treatment of lumbar disc herniation by artificial intelligence will evolve towards standardization,efficiency,and precision through continuous optimization of algorithms and the professional development of data platforms.

[This corrects the article DOI: 10.1016/j.apsb.2024.04.021.].

BACKGROUND:Lumbar fixed-point rotation operation needs collaborative operation of the doctor's hands,and outputs rotation and thumb thrust.Lumbar disc herniation can be treated through disc displacement and adjusting stress distribution.However,the mechanical effects of thumb thrust and the biomechanical effects of loading direction on manipulative effects remain unclear. OBJECTIVE:To compare the biomechanical difference of lumbar fixed-point rotation manipulation for treating lumbar disc herniation under different thrust directions. METHODS:The L3-5 normal three-dimensional finite element model was constructed and validity was verified.According to the intervertebral disc degeneration Pfirrmann grade,intervertebral disc degeneration was simulated by modifying the L4/5 intervertebral space height,the volume of the nucleus pulposus,as well as the material parameters of the annulus fibrosus,nucleus pulposus,and ligament.Finally,the pathological model of L4/5 moderate disc degeneration with left para-central herniation was constructed,and then the pathological models were used as research objects.Simulation technique:spinning to the right;taking the condition on changing the direction of the thumb thrust to establish three modes of operation(M1:thumb push to the left;M2:thumb push to the right;M3:no thrust push).The protrusion displacement and the disc stress,and the stress and strain of the facet joint cartilage were compared in the three operating modes. RESULTS AND CONCLUSION:(1)Maximum displacement value of L4/5 disc herniation:displacement was 2.672 3 mm for M1,1.156 1 mm for M2,1.826 4 mm for M3,M1＞M3＞M2.(2)The maximum Von Mises stress of L4/5 discs was 1.846 7 MPa for M1,0.419 0 MPa for M2,and 1.257 9 MPa for M3,M1＞M3＞M2.(3)L4/5 bilateral small cartilage produced different degrees of contact stress changes:It was 0.485 5 MPa for M1,0.026 7 MPa for M2,and 0.441 4 MPa for M3,M1＞M3＞M2.Right cartilage contact force was 0.000 5 MPa for M1,0.025 9 MPa for M2,and 0.001 3 MPa for M3,M2＞M3＞M1;the left greater than the right,M1 had the highest value;cartilage strain was consistent with contact stress changes.(4)Different operation modes will have some biomechanical influences on the diseased intervertebral disc and accessory structure.The M1 operation mode can maximize the displacement of protrusion,disc stress and left joint cartilage contact,which can better promote disc displacement,balance stress distribution and reduce facet joint disorder,so the operation is better.

Objective:To explore the clinical effect of radial extracorporeal shockwave therapy on delayed union of forearm fractures in children with ultrasonic guidance.Methods:A retrospective analysis of information on 18 children with delayed forearm fracture union who received ultrasonic guided extracorporeal shockwave therapy from February 2018 to June 2019 was conducted. Among them, there were 14 males and 4 females; Age: 9.44±3.75 years (range, 3-15 years); All the children were closed forearm fractures, including 13 cases of ulna and 5 cases of radius. Initial fixation methods: intramedullary nails fixation in 8 cases, Kirschner wire fixation in 4 cases, steel plate fixation in 2 cases, external fixation in 2 cases, conservative treatment in 2 cases; The classification of fracture nonunion were: 14 cases of hypertrophy, 4 cases of atrophy; The course of disease was 4.0 (3.0, 6.0) months. The front and lateral X-ray films of the affected side forearm were taken before treatment, 3 months and 6 months after treatment. The scores of callus condition were performed using Lane-Sandhu X-ray scoring standard and Fernandez-Esteve X-ray evaluation standard of callus grade.Results:All children completed treatment and were followed up for 6 months. The bone healing standard was the disappearance of the fracture line shown by anterior and lateral X-ray films. Within 6 months after treatment, 11 patients got bone union. The healing rate was 61.11% (11/18). The average of Lane-Sandhu X-ray scores before treatment, 3 months and 6 months after treatment were 3.0 (1.0, 4.0), 6.0 (4.0, 8.0) and 10.0 (5.0, 12.0), respectively, there were statistically significant differences in pairwise comparisons at each time point. And the average scores of Fernandez-Esteve X-ray evaluation standard for callus grade were 1.0 (1.0, 2.0), 3.0 (2.0, 4.0), and 4.0 (3.0, 4.0), respectively, there were statistically significant differences from 3 months and 6 months after treatment to preoperative group, while there was no statistically significant difference between 3 months and 6 months after treatment. The mixed effects model analysis results showed that only the Lane Sandhu X-ray score and Fernandez Esteve X-ray standard score of callus grade at different follow-up time points were significantly different. There was no statistically significant difference in age, gender, number of shockwave treatments and interval time from the first treatment after the initial fixation.Conclusion:The radial extracorporeal shockwave therapy can effectively treat the delayed healing of forearm fractures in children; the X-ray score has been significantly improved. The treatment is highly accepted by children and their parents, and can be used as one of the methods to treat delayed healing of fractures in children in the future.

Objective:To study the a feasibility of modified pharyngoplasty in the treatment of velopharyngeal insufficiency.Methods:30 cases of velopharyngeal insufficiency were treated with a modified pharyngoplasty followed by speech training.The effects of the treatment were evaluated by blowing test (BT),chinese articulation inteligibility test (CIT) and hypernasality observation(HO).Results:Before and after treatment BT was 1～26 and 8～40,CIT(%) 20～60 and 47～82,HO(cases) 30 and 3,respectively.Conclusions:The modified pharyngoplastic surgery is feasible in the treatment of velophayngeal insufficiency.