<?xml version="1.0" encoding="UTF-8"?>
<article article-type="Research Article">
  <front>
    <journal-meta>
      <journal-id journal-id-type="publisher">journal-of-information-and-optimization-sciences</journal-id>
      <journal-title-group>
        <journal-title>Journal of Information and Optimization Sciences</journal-title>
      </journal-title-group>
      <issn publication-format="electronic">2169-0103</issn>
      <issn publication-format="print">0252-2667</issn>
      <publisher>
        <publisher-name>Taru Publications</publisher-name>
      </publisher>
    </journal-meta>
    <article-meta>
      <article-id pub-id-type="doi">10.47974/JIOS-2251</article-id>
      <title-group>
        <article-title>Comparative evaluation of transformer and CNN architectures for breast cancer detection across histopathology, mammography, and ultrasound</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author" corresp="yes">
          <name>
            <surname>Ghuge</surname>
            <given-names>Kalyani</given-names>
          </name>
          <aff>School of Computing Science &amp; Engineering, Kothrikalan, VIT Bhopal University, Sehore, Madhya Pradesh, 466114, India</aff>
        </contrib>
        <contrib contrib-type="author">
          <name>
            <surname>Saravanan</surname>
            <given-names>D.</given-names>
          </name>
          <aff>School of Computing Science &amp; Engineering, Kothrikalan, VIT Bhopal University, Sehore, Madhya Pradesh, 466114, India</aff>
        </contrib>
      </contrib-group>
      <volume>47</volume>
      <issue>5-A</issue>
      <fpage>1613</fpage>
      <lpage>1621</lpage>
      <pub-date date-type="pub">
        <day>23</day>
        <month>04</month>
        <year>2026</year>
      </pub-date>
      <abstract>
        <p>Automated analysis of images in various imaging modalities are beneficial in breast cancer screening and diagnosis. This paper compares four deep learning models Vision Transformer (ViT), EfficientNetV2 with CBAM, ConvMixer, and MobileNetV4 on three publicly available datasets of breast imaging (histopathology, mammography, ultrasound). We measure accuracy, recall, F1-score, computational cost, and suitability to modality. To underline the focus on the reproducibility and modality-specific analysis, the methodology underlines concise descriptions of every model in terms of mathematical intuition, architecture illustration, algorithmic steps. We have found that attention-augmented CNNs (CBAM-EfficientNetV2) perform well in terms of accuracy and robustness across modalities and MobileNetV4 has a high efficiency-accuracy trade-off to be deployed on resource-constrained devices.</p>
      </abstract>
      <kwd-group>
        <kwd>Breast cancer detection</kwd>
        <kwd>Vision transformer</kwd>
        <kwd>CBAM-EfficientNetV2</kwd>
        <kwd>ConvMixer</kwd>
        <kwd>MobileNetV4</kwd>
        <kwd>Histopathology</kwd>
        <kwd>Mammography</kwd>
        <kwd>Ultrasound</kwd>
      </kwd-group>
      <custom-meta-group>
        <custom-meta>
          <meta-name>access</meta-name>
          <meta-value>open</meta-value>
        </custom-meta>
        <custom-meta>
          <meta-name>retracted</meta-name>
          <meta-value>no</meta-value>
        </custom-meta>
      </custom-meta-group>
    </article-meta>
  </front>
</article>
