Scholarlink

<p>Magnetic Resonance Imaging (MRI) is the most commonly implemented alternative for diagnosing intracranial tumors. To perform classification efficiently, conventional alternatives should be replaced with rule-based approaches. In this study, a hybrid approach utilizing distinct Spike neural networks (SNNs) for diagnosis of intracranial tumors is proposed. It comprises Spike timing-dependent plasticity and backpropagation neural networks (STDP + BPNN). The proposed SNNs resemble the networks that have three convolutional layers (Conv1, Conv2, and Conv3), three pooling layers (Pool1, Pool2, and Pool3), and a DOG (Difference of Gaussians) encoding layer. Using the intensity-to-latency coding technique, the pre-processed tumor pictures are transformed into spikes by applying the DOG filter. Then, the DOG filter’s output spikes are sorted into a few consecutive time steps to get analyzed by the convolutional layer. The spike-timing-dependent plasticity (STDP) learning rule is applied that adjusts synaptic weights based on the timing of spikes between pre- and post-synaptic neurons to enhance its performance and applicability in different contexts. Simulation results show that the proposed approach achieved maximum average accuracy of about 97.65% with the highest AUC score of 98.38 and the lowest computational score (167.86&#xa0;s) thus surpassing existing state-of-the-art techniques. Therefore, we conclude that the proposed distinct learning rule approach can be employed as operational tool for futuristic medical images classification purposes.</p>

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

A novel deep learning rule-based spike neural network (SNN) classification approach for diagnosis of intracranial tumors

  • 期刊论文

  • 发表日期：

    2025年10月6日

查看全文
  • Pijush Dutta,
  • Anubrata Mondal,
  • Rahul Vadisetty,
  • Anand Polamarasetti,
  • Raviteja Guntupalli,
  • Sateesh Kumar Rongali

摘要

Magnetic Resonance Imaging (MRI) is the most commonly implemented alternative for diagnosing intracranial tumors. To perform classification efficiently, conventional alternatives should be replaced with rule-based approaches. In this study, a hybrid approach utilizing distinct Spike neural networks (SNNs) for diagnosis of intracranial tumors is proposed. It comprises Spike timing-dependent plasticity and backpropagation neural networks (STDP + BPNN). The proposed SNNs resemble the networks that have three convolutional layers (Conv1, Conv2, and Conv3), three pooling layers (Pool1, Pool2, and Pool3), and a DOG (Difference of Gaussians) encoding layer. Using the intensity-to-latency coding technique, the pre-processed tumor pictures are transformed into spikes by applying the DOG filter. Then, the DOG filter’s output spikes are sorted into a few consecutive time steps to get analyzed by the convolutional layer. The spike-timing-dependent plasticity (STDP) learning rule is applied that adjusts synaptic weights based on the timing of spikes between pre- and post-synaptic neurons to enhance its performance and applicability in different contexts. Simulation results show that the proposed approach achieved maximum average accuracy of about 97.65% with the highest AUC score of 98.38 and the lowest computational score (167.86 s) thus surpassing existing state-of-the-art techniques. Therefore, we conclude that the proposed distinct learning rule approach can be employed as operational tool for futuristic medical images classification purposes.