TY  - JOUR
AU  - Naseer, Fahad
AU  - Jamal, Ahmad
AU  - Tauseef, Fatima
AU  - Nasim, Muhammad Fawad
PY  - 2026
DA  - 2026/09/30
TI  - AI-Driven Mammography for Early Breast Cancer Detection: A Three-Tier Framework Integrating CNN-ViT Hybrid Diagnosis, Explainable AI, and Robotic-Assisted Biopsy
JO  - Journal of Artificial Intelligence in Bioinformatics
T2  - Journal of Artificial Intelligence in Bioinformatics
JF  - Journal of Artificial Intelligence in Bioinformatics
VL  - 2
IS  - 2
SP  - 68
EP  - 79
DO  - 10.62762/JAIB.2026.442596
UR  - https://www.icck.org/article/abs/JAIB.2026.442596
KW  - breast cancer detection
KW  - mammography
KW  - CNN-ViT hybrid
KW  - explainable AI (XAI)
KW  - medical robotics
KW  - radiogenomics
AB  - Breast cancer remains the leading malignancy among women worldwide, while conventional mammography screening is limited by inter-reader variability, radiologist workload, and false-negative diagnoses. This study evaluates a three-tier intelligent framework integrating AI classification, explainable reporting, and robot-assisted biopsy guidance. A hypothetical cohort of 3,542 mammographic cases was constructed from published distributions of three benchmark datasets and evaluated using a 60/20/20 train–validation–test split with stratified 5-fold cross-validation. The framework comprises a CNN–ViT hybrid combining ResNet-50 feature extraction with Vision Transformer attention, a triple-modality explainability layer integrating Grad-CAM, SHAP, and LIME, and a Medical Robotics subsystem for virtual biopsy planning. GAN-based augmentation was applied to address class imbalance. The CNN–ViT achieved an AUC of 0.947, sensitivity of 96.2%, specificity of 93.8%, and F1-score of 95.1%, significantly outperforming the CNN-only baseline (AUC 0.913; p < 0.001) and reducing the false-negative rate by approximately 75% relative (11.6 percentage points absolute) to radiologist-only reading. The explainability layer achieved 88.2% radiologist acceptance and reduced review time by 28.3% (p = 0.003). Robotic needle placement accuracy reached 1.3 ± 0.4 mm, while radiogenomic classifiers improved molecular subtype prediction by 5.9%–8.3%. These findings demonstrate the framework’s potential to improve diagnostic accuracy, interpretability, procedural precision, and molecular prediction, supporting further prospective multi-institutional validation.
SN  - 3068-7535
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Naseer2026AIDriven,
  author = {Fahad Naseer and Ahmad Jamal and Fatima Tauseef and Muhammad Fawad Nasim},
  title = {AI-Driven Mammography for Early Breast Cancer Detection: A Three-Tier Framework Integrating CNN-ViT Hybrid Diagnosis, Explainable AI, and Robotic-Assisted Biopsy},
  journal = {Journal of Artificial Intelligence in Bioinformatics},
  year = {2026},
  volume = {2},
  number = {2},
  pages = {68-79},
  doi = {10.62762/JAIB.2026.442596},
  url = {https://www.icck.org/article/abs/JAIB.2026.442596},
  abstract = {Breast cancer remains the leading malignancy among women worldwide, while conventional mammography screening is limited by inter-reader variability, radiologist workload, and false-negative diagnoses. This study evaluates a three-tier intelligent framework integrating AI classification, explainable reporting, and robot-assisted biopsy guidance. A hypothetical cohort of 3,542 mammographic cases was constructed from published distributions of three benchmark datasets and evaluated using a 60/20/20 train–validation–test split with stratified 5-fold cross-validation. The framework comprises a CNN–ViT hybrid combining ResNet-50 feature extraction with Vision Transformer attention, a triple-modality explainability layer integrating Grad-CAM, SHAP, and LIME, and a Medical Robotics subsystem for virtual biopsy planning. GAN-based augmentation was applied to address class imbalance. The CNN–ViT achieved an AUC of 0.947, sensitivity of 96.2\%, specificity of 93.8\%, and F1-score of 95.1\%, significantly outperforming the CNN-only baseline (AUC 0.913; p < 0.001) and reducing the false-negative rate by approximately 75\% relative (11.6 percentage points absolute) to radiologist-only reading. The explainability layer achieved 88.2\% radiologist acceptance and reduced review time by 28.3\% (p = 0.003). Robotic needle placement accuracy reached 1.3 ± 0.4 mm, while radiogenomic classifiers improved molecular subtype prediction by 5.9\%–8.3\%. These findings demonstrate the framework’s potential to improve diagnostic accuracy, interpretability, procedural precision, and molecular prediction, supporting further prospective multi-institutional validation.},
  keywords = {breast cancer detection, mammography, CNN-ViT hybrid, explainable AI (XAI), medical robotics, radiogenomics},
  issn = {3068-7535},
  publisher = {Institute of Central Computation and Knowledge}
}