<?xml version="1.0" encoding="UTF-8"?>



<records>

  <record>
    <language>eng</language>
          <publisher>Oriental Scientific Publishing Company</publisher>
        <journalTitle>Biomedical and Pharmacology Journal</journalTitle>
          <issn>0974-6242</issn>
            <publicationDate>2026-11-03</publicationDate>
    
        <volume>19</volume>
        <issue>4</issue>

 
    <startPage></startPage>
    <endPage></endPage>

	    <publisherRecordId>73910</publisherRecordId>
    <documentType>article</documentType>
    <title language="eng">A Multi-Scale Feature Fusion Approach with Attention for Robust Ovarian Tumor Diagnosis</title>

    <authors>
	 


      <author>
       <name>Ankita Nainwal</name>

 
		
	<affiliationId>1</affiliationId>
      </author>
    

	 


      <author>
       <name>Vidit Kumar </name>


		
	<affiliationId>1</affiliationId>

      </author>
    

	 


      <author>
       <name>Bhaskar Pant </name>

		
	<affiliationId>1</affiliationId>
      </author>
    

	


	


	
    </authors>
    
	    <affiliationsList>
	    
		
		<affiliationName affiliationId="1">Department of Computer Science and Engineering, Graphic Era Deemed to University, Dehradun, India</affiliationName>
    

		
		
		
		
		
	  </affiliationsList>






    <abstract language="eng">Ovarian cancer ranks high among diagnosed cancers in women across the globe. Early identification plays a key role, since timely intervention often shapes outcomes more than late-stage efforts. Because of this, screening strategies continue evolving under clinical pressure. Tools like MRI, CT scans, and ultrasound help visualize abnormalities - each bringing distinct advantages to diagnosis. Within this space, our work focuses on improving how ultrasound data classifies tumors, applying deep learning models to refine accuracy beyond traditional interpretation. Looking at how four well-known deep learning models - VGG16, ResNet50, EfficientNet, and DenseNet201 - performed on the MMOTU dataset was where this work began. ResNet50 stood out due to its stable outcomes, leading later efforts to center around it. With an aim to boost its ability to interpret data, a component called the Multi-Scale Feature Enhancement Block (MSFEB) was integrated; this helped uncover more nuanced patterns across images. The MSFEB block consists of three parallel branches: two branches employ stacked 3×3 convolutional layers to capture fine and intermediate texture details, while the third branch uses a 5×5 convolution to model broader contextual information. The features from all branches are fused and followed by a Global Average Pooling (GAP) layer to generate compact and discriminative representations. Separately, we evaluated the Convolutional Block Attention Module (CBAM), which further enhanced the network by focusing on salient regions. Finally, MSFEB was fused with CBAM, and the combined architecture achieved the highest classification accuracy of 76.55% and a top-2 accuracy of 86.99%, confirming its robustness for ovarian cancer classification.</abstract>

    <fullTextUrl format="html">https://biomedpharmajournal.org/vol19no4/a-multi-scale-feature-fusion-approach-with-attention-for-robust-ovarian-tumor-diagnosis/</fullTextUrl>

<keywords language="eng">

      
        <keyword>A Computer-Aided Design (CAD)</keyword>
      

      
        <keyword> Convolutional Block Attention Module (CBAM)</keyword>
      

      
        <keyword> Convolutional Neural Network (CNN)</keyword>
      

      
        <keyword> Computed Tomography (CT)</keyword>
      

      
        <keyword> Magnetic Resonance Imaging (MRIs)</keyword>
      

      
        <keyword> Multi-Scale Feature Enhancement Block (MSFEB)</keyword>
      
</keywords>
  </record>
</records>