Scholarlink

<p>Common bone cancer predominantly affecting young individuals is osteosarcoma. Particularly considering tissue samples stained with hematoxylin and eosin (H&amp;E), pathologists find it somewhat difficult to categorize this malignancy. These cells vary greatly, seem identical, and sometimes contain noise; so, precise diagnosis of them is challenging. This paper presents a mixed technique utilizing a series of whole slide images (WSI) to better identify osteosarcoma tumors into three categories: non-tumor, necrosis, and live tumor. We employ transfer learning to train five well-known CNN models after using many stages to prepare the data. Every model is optimized using several variables to raise its performance. We thoroughly assess our hybrid model using feature extraction from convolutional and pooling layers. For jobs with two or more categories, we further assess the model using five-fold cross-valuation and employ an upgraded CNN classifier. With a binary accuracy of 99.4% and a multiclass accuracy of 95.2%, the outcomes reveal outstanding success. Showing it may be used in hospitals to diagnose osteosarcoma, the suggested model is far superior to present approaches. Effective in accurately diagnosing osteosarcoma from small sample numbers, we present a novel RCNN model combining CNN with bidirectional LSTM. Though it suffers with overfitting, variances within the same class, similarities across various classes, cluttered backgrounds, unequal cell shapes, and noise in H&amp;E stained pictures, our model shows significant potential. More helpful for clinical studies, we created a web app using FastAPI providing real-time predictions.</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

Enhanced Osteosarcoma Diagnosis Through Combining CNN-BiLSTM Architectures and Multilayer Perceptrons for Histopathological Image Analysis

  • 期刊论文

  • 发表日期：

    2026年5月14日

查看全文
  • J. Sindudevi,
  • M. G. Kavitha

摘要

Common bone cancer predominantly affecting young individuals is osteosarcoma. Particularly considering tissue samples stained with hematoxylin and eosin (H&E), pathologists find it somewhat difficult to categorize this malignancy. These cells vary greatly, seem identical, and sometimes contain noise; so, precise diagnosis of them is challenging. This paper presents a mixed technique utilizing a series of whole slide images (WSI) to better identify osteosarcoma tumors into three categories: non-tumor, necrosis, and live tumor. We employ transfer learning to train five well-known CNN models after using many stages to prepare the data. Every model is optimized using several variables to raise its performance. We thoroughly assess our hybrid model using feature extraction from convolutional and pooling layers. For jobs with two or more categories, we further assess the model using five-fold cross-valuation and employ an upgraded CNN classifier. With a binary accuracy of 99.4% and a multiclass accuracy of 95.2%, the outcomes reveal outstanding success. Showing it may be used in hospitals to diagnose osteosarcoma, the suggested model is far superior to present approaches. Effective in accurately diagnosing osteosarcoma from small sample numbers, we present a novel RCNN model combining CNN with bidirectional LSTM. Though it suffers with overfitting, variances within the same class, similarities across various classes, cluttered backgrounds, unequal cell shapes, and noise in H&E stained pictures, our model shows significant potential. More helpful for clinical studies, we created a web app using FastAPI providing real-time predictions.