Scholarlink

Differential assessment of subcutaneous adipose tissue (SAT) and intermuscular adipose tissue (IMAT), two forms of muscle fat, is necessary for studying manifestations of ageing, muscle atrophy, sarcopenia, obesity and metabolic diseases such as diabetes. The discrimination of SAT and IMAT by ultrasonic measurements is difficult due to their complex influence. In the present study, machine-learning algorithms applied to key parameters extracted from ultrasound propagation signals obtained in simplified tissue models (phantoms) were investigated. The acoustical phantoms of muscle tissue were made of gelatin with oil simulating fat layers (SAT) and inner inclusions (IMAT). SAT and IMAT contents varied from zero to 50% with a step 12.5%. A specialised recurrent neural network (RNN) architecture, the long short-term memory (LSTM) method is used in this paper and was used as the main method in the experiments. The result of SAT and IMAT evaluation of objects with an error of no more than 3% in 95% of cases.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Application of Neural Networks to Ultrasonic Data for Discrimination of Fat Types in Muscle Tissue Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jegors Lukjanovs,
  • Aleksandrs Sisojevs,
  • Alexey Tatarinov,
  • Tamara Laimiņa

摘要

Differential assessment of subcutaneous adipose tissue (SAT) and intermuscular adipose tissue (IMAT), two forms of muscle fat, is necessary for studying manifestations of ageing, muscle atrophy, sarcopenia, obesity and metabolic diseases such as diabetes. The discrimination of SAT and IMAT by ultrasonic measurements is difficult due to their complex influence. In the present study, machine-learning algorithms applied to key parameters extracted from ultrasound propagation signals obtained in simplified tissue models (phantoms) were investigated. The acoustical phantoms of muscle tissue were made of gelatin with oil simulating fat layers (SAT) and inner inclusions (IMAT). SAT and IMAT contents varied from zero to 50% with a step 12.5%. A specialised recurrent neural network (RNN) architecture, the long short-term memory (LSTM) method is used in this paper and was used as the main method in the experiments. The result of SAT and IMAT evaluation of objects with an error of no more than 3% in 95% of cases.