Reshaping Breast Cancer Care: AI-Driven Advances in Detection, Diagnosis, and Prognosis

Written by Promit Ghosal, Prosenjit Kundu & Ayowole Delegan

In recent years, artificial intelligence (AI) has increasingly been integrated into various fields of medicine, including oncology and, more specifically, breast cancer research and care. This review explores four key areas where AI and breast cancer intersect:

• AI-driven prediction for breast cancer targets integrating multimodal data
• AI Histopathology driven breast cancer characterization
• AI-driven treatment planning in breast cancer
• AI-based prediction of survival due to treatment response

Better Healthcare with AI (bHAI) Logo

AI-driven prediction for breast cancer targets integrating multimodal data

The Breast Imaging Reporting and Data System (BI-RADS) is a widely used tool for assessing breast cancer risk based on imaging findings like mammography, ultrasound, and MRI. Endorsed by the American Cancer Society (ACS) [1] for its standardized reporting, BI-RADS improves communication among healthcare providers but relies on radiologist expertise, introducing some subjectivity. It is qualitative, focusing solely on imaging findings, and does not incorporate factors like reproductive history, polygenic risk scores, or histopathological data. Over-reliance on imaging findings can lead to unnecessary biopsies or missed malignancies, especially in dense breast tissue where imaging sensitivity is reduced. To enhance breast cancer risk assessment, BI-RADS can be complemented with AI-based tools that integrate diverse data types.

In general, AI-driven tools for breast cancer detection are advancing rapidly, utilizing machine learning (ML), multimodal data integration, and innovative deployment strategies for more comprehensive breast cancer risk assessment. Recent models integrate diverse data types to enhance predictive accuracy. Hussain et al. [2] reviewed deep learning (DL) and traditional methods that incorporate imaging (e.g., mammograms, ultrasounds, MRIs, PET scans) and non-imaging data such as genomic features, polygenic risk scores, and lifestyle factors. One notable example is the Mirai model, which uses mammographic density-based DL to identify high-risk patients effectively across datasets, addressing missing data on traditional risk factors and demonstrating robust performance across populations.

Nakach et al. [3] provided a review of multimodal DL (MMDL) techniques for breast cancer classification. Their findings revealed that most studies integrated clinical data with imaging or genes, though only a small fraction combined all three modalities. The research spans diverse targets, including survival prediction (44%), binary classification of malignant versus benign tumors (21%), molecular subtype classification (13%), and recurrence prediction (8%).

The focus of MMDL tools is shifting from merely enhancing predictive accuracy to improving interpretability, which is essential for clinical adoption. A recent study by Qian et al. [4] introduced the BMU-Net model which integrates mammograms, ultrasound images, and clinical metadata, leveraging pre-trained weights from the Mirai model. BMU-Net demonstrated superior performance in pathology-level diagnosis and provided heatmaps to guide clinicians in evaluating influential regions, enhancing trust in AI predictions.

Explainable AI (XAI) techniques, such as Shapley additive explanations (SHAP), Local Interpretable Model-agnostic Explanations (LIME), and Gradient-weighted Class Activation Mapping (GRAD-CAM), have been increasingly employed to improve transparency. A review by Amirehsan Ghasemi et al. [5] identified SHAP as the most popular model-agnostic XAI technique in breast cancer research. Additionally, emerging tools like generative models augmented with GRAD-CAM or attention CAM as highlighted by Faseela Abdullakutty et al. [6] are driving progress in multimodal XAI.

These advancements promise improved accuracy and explainability, fostering greater confidence among clinicians and patients.

AI histopathology-driven breast cancer characterization

Artificial intelligence (AI) has revolutionized breast cancer histopathology by offering powerful tools for diagnosis, prognosis, and treatment planning. Integrating machine learning (ML) and deep learning (DL) algorithms with digital pathology has significantly enhanced diagnostic accuracy.

AI-driven systems, such as the GALEN algorithm, excel at analyzing whole slide images (WSIs) to detect breast lesions, including invasive carcinoma and ductal carcinoma in situ (DCIS). They have impressive performance metrics (AUC (area under the curve) of 0.99 and 0.98, respectively) [7].

AI also shows promise in histological subtyping of breast cancer, achieving high accuracy (93.2% on the BreaKHis dataset) in identifying ductal, lobular, and rarer subtypes [7,8]. These advancements underscore AI’s transformative potential in improving the precision of breast cancer diagnosis. Beyond detection, AI tools have demonstrated value in prognostic assessments and treatment planning. In evaluating tumor-infiltrating lymphocytes (TILs), critical prognostic markers in breast cancer, AI-based methods offer precise and automated scoring capabilities [9].

Similarly, Ki-67 proliferation index assessments, crucial for guiding treatment decisions, benefit from AI-enhanced accuracy and standardization, significantly reducing errors in clinical evaluation (2.1% error with AI versus 5.9% without AI) [10]. Furthermore, AI has been used to predict responses to neoadjuvant chemotherapy (NAC) by analyzing pathomic features from biopsy samples, achieving high predictive performance (AUC of 0.90) compared to models using only clinical data [11]. AI’s integration into biomarker assessments, such as predicting biomarkers directly from H&E slides, offers a pathway to streamlined diagnostics and reduced costs [12].

As AI applications in histopathology continue to evolve, commercial platforms like PathcoreFlow™ are being integrated into clinical workflows, providing tools for automated scoring and visual analyses [13]. These advancements, while promising, still face challenges, including ensuring the generalizability of AI models across diverse datasets and addressing the “black-box” nature of some algorithms [13].

Ongoing research aims to refine these models and extend their capabilities. With continued innovation and validation, AI-driven histopathology is poised to enhance diagnostic accuracy, improve prognostic evaluations, and support personalized treatment strategies for breast cancer, ultimately transforming patient care.

Shapley Additive Explanations — most promiment XAI technique

AI-driven treatment planning in breast cancer

AI-driven advancements in tumor definition and treatment planning have significantly transformed radiation therapy for breast cancer, enhancing both efficiency and accuracy. Automated contouring, a core component of these advancements, has greatly reduced the time-consuming task of manually delineating organs at risk (OARs) and target volumes, as demonstrated in studies focusing on radiotherapy planning for breast cancer patients [14,15].

For instance, convolutional neural network (CNN)-based models have shown remarkable promise in generating clinically acceptable radiotherapy plans for locally advanced breast cancer. By leveraging 2D distance maps of relevant regions of interest as input, these models have achieved dose predictions that meet all clinical goals, with minimal differences (≤0.5 Gy) compared to manually created plans, maintaining consistent plan quality while significantly speeding up the planning process [16,17].

Beyond time efficiency, AI-based planning systems have demonstrated improvements in plan quality and adaptability. Research indicates that AI-driven systems can reduce total planning time from an average of 163 ± 97 minutes for manual plans to just 33 ± 5 minutes for automated plans, with user interaction time reduced from approximately 130 minutes to only 5 minutes [18].

Additionally, these systems have enhanced planning target volume (PTV) coverage without compromising OAR doses, particularly in complex scenarios like accelerated partial breast irradiation using volumetric modulated arc therapy (VMAT) [18]. The adaptability of AI systems extends to real-time adjustments in adaptive radiotherapy, where AI-driven platforms analyze daily imaging data to modify treatment plans based on anatomical changes or tumor size variations, ensuring precision and minimizing radiation exposure to healthy tissues [19].

As the field evolves, ongoing efforts aim to improve the generalizability of AI models and address challenges such as the “black-box” nature of algorithms, paving the way for further advancements in breast cancer radiation therapy [20,21,22].

AI-based prediction of survival due to treatment response

Breast cancer prognosis is often not favorable, stressing the need for early metastasis prediction. As with other aspects of breast cancer management, AI has found significant use in predicting recurrence risk and survival amongst post-operative patients i.e., Disease Free Survival (DFS).

Wang et al. [23], attempted to predict DFS in nonmetastatic breast cancer using a model (Deep Learning Clinical Medicine-Based Pathological Gene Multimodal or DeepClinMed-PGM) integrating molecular data, pathological slides, and clinical information to improve understanding and prognosis. Combining multimodal data into the DeepClinMed-PGM model significantly enhanced its predictive performance for DFS (1-, 3- and 5- years), demonstrating robust discriminative capabilities across training, validation, and external cohorts. Their work reflects the ongoing change of direction toward using AI algorithms to perform difficult medical field tasks.

Neoadjuvant therapy (NAT), a treatment given before the primary intervention, is standard for patients with inoperable but resectable breast cancer. Pathological Complete Response (pCR), which could be described as the absence of residual invasive cancer cells in histopathological specimens of breast and axillary lymph nodes after NAT (ypT0/is, ypN0), is associated with favorable survival outcomes [24]. While the NAT procedure shows little variation across settings, some patients do not respond well, suggesting that, despite its toxicity, the treatment may offer no real benefits for them. In a scenario such as this, it is possible to predict pCR in such patients before administration of NAT in a way that helps limit exposure to invasive treatments using deep learning techniques. Gao et al. [24], built a multi-modal response prediction (MRP) system tailored to specific response prediction of neoadjuvant therapy in breast cancer patients. The MRP system utilized longitudinal data throughout the NAT process within real clinical contexts, integrating multi-modal data sources including imaging, histology, personal factors, clinical data, and therapy details of eligible breast cancer participants.

The outcome of this study by Gao [24] revealed that the:

• MRP model demonstrates a superior ability to accurately predict pathological complete response
• MRP presented comparable robustness and generalizability to breast radiologists, significantly outperforming humans in pCR prediction on the Pre-NAT phase
• MRP shows its interpretability in design for training/inference

Future direction & BHAI’s approach

Despite advancements in breast cancer research, integrating diverse data types, such as polygenic risk scores (PRS), imaging data (e.g., mammograms), and clinical information, remains a promising but underutilized approach. Recent studies suggest that such integration can enhance risk prediction and patient stratification [25,26], though limited genetic data availability, especially in non-European populations [27], poses challenges. Developing multi-ethnic cohorts and interpretable machine learning models to handle heterogeneous data [28,29] is essential for addressing these gaps and helping provide more personalized treatment decisions [30,31].

Combining advanced imaging techniques like radiomics with genetic data shows potential for improving early detection [25,32]. Hybrid approaches, merging radiologist assessments (e.g., BI-RADS) with AI-based prediction tools, have demonstrated significant benefits. For instance, Germany’s PRAIM study [33] reported a 17.6% improvement in detection rates and reduced radiologists’ workloads using AI-integrated screening.

Future research should prioritize standardizing data integration protocols [27,34] and exploring the interplay of imaging, genetic, environmental, and lifestyle factors in cancer progression [30,35] using interpretable multi-modal AI techniques while incorporating emerging risk factors from each modality.

At BHAI, we remain committed to advancing these efforts in 2025, aiming to improve patient outcomes and address healthcare disparities.

Bibliography

1. https://www.cancer.org/cancer/types/breast-cancer/screening-tests-and-early-detection/mammograms/understanding-your-mammogram-report.html?
2. Hussain S, Ali M, Naseem U, Nezhadmoghadam F, Jatoi MA, Gulliver TA, Tamez-Peña JG. Breast cancer risk prediction using machine learning: a systematic review. Front Oncol. 2024 Mar 20;14:1343627. doi: 10.3389/fonc.2024.1343627. PMID: 38571502; PMCID: PMC10987819.https://pmc.ncbi.nlm.nih.gov/articles/PMC10987819/
3. Nakach, FZ., Idri, A. & Goceri, E. A comprehensive investigation of multimodal deep learning fusion strategies for breast cancer classification. Artif Intell Rev 57, 327 (2024). https://doi.org/10.1007/s10462-024-10984-z
4. Qian, X., Pei, J., Han, C. et al. A multimodal machine learning model for the stratification of breast cancer risk. Nat. Biomed. Eng (2024). https://doi.org/10.1038/s41551-024-01302-7
5. Ghasemi A, Hashtarkhani S, Schwartz DL, Shaban-Nejad A. Explainable artificial intelligence in breast cancer detection and risk prediction: A systematic scoping review. Cancer Innov. 2024 Jul 3;3(5):e136. https://doi.org/10.1002/cai2.136
6. Abdullakutty F, Akbari Y, Al-Maadeed S, Bouridane A, Talaat IM and Hamoudi R (2024) Histopathology in focus: a review on explainable multi-modal approaches for breast cancer diagnosis. Front. Med. 11:1450103. https://doi.org/10.3389/fmed.2024.1450103
7. Ivanova M, Pescia C, Trapani D, Venetis K, Frascarelli C, Mane r4E, Cursano G, Sajjadi E, Scatena C, Cerbelli B, d’Amati G, Porta FM, Guerini-Rocco E, Criscitiello C, Curigliano G, Fusco N. Early Breast Cancer Risk Assessment: Integrating Histopathology with Artificial Intelligence. Cancers (Basel). 2024 May 23;16(11):1981. doi:10.3390/cancers16111981. PMID: 38893102; PMCID: PMC11171409.
8. Priya C V L, V G B, B R V, Ramachandran S. Deep learning approaches for breast cancer detection in histopathology images: A review. Cancer Biomark. 2024;40(1):1–25. doi:10.3233/CBM-230251. PMID: 38517775; PMCID: PMC11191493.
9. Chan, R.C., To, C.K.C., Cheng, K.C.T., Yoshikazu, T., Yan, L.L.A. and Tse, G.M. (2023), Artificial intelligence in breast cancer histopathology. Histopathology, 82: 198–210. https://doi.org/10.1111/his.14820
10. Dy, A., Nguyen, NN.J., Meyer, J. et al. AI improves accuracy, agreement and efficiency of pathologists for Ki67 assessments in breast cancer. Sci Rep 14, 1283 (2024). https://doi.org/10.1038/s41598-024-51723-2
11. Saednia, K., Lagree, A., Alera, M.A. et al. Quantitative digital histopathology and machine learning to predict pathological complete response to chemotherapy in breast cancer patients using pre-treatment tumor biopsies. Sci Rep 12, 9690 (2022). https://doi.org/10.1038/s41598-022-13917-4
12. Ibrahim A, Gamble P, Jaroensri R, Abdelsamea MM, Mermel CH, Chen PC, Rakha EA. Artificial intelligence in digital breast pathology: Techniques and applications. Breast. 2020 Feb;49:267–273. doi:10.1016/j.breast.2019.12.007. Epub 2019 Dec 19. PMID: 31935669; PMCID: PMC7375550.
13. Ahn JS, Shin S, Yang SA, Park EK, Kim KH, Cho SI, Ock CY, Kim S. Artificial Intelligence in Breast Cancer Diagnosis and Personalized Medicine. J Breast Cancer. 2023 Oct;26(5):405–435. https://doi.org/10.4048/jbc.2023.26.e45
14. van de Sande D, Sharabiani M, Bluemink H, Kneepkens E, Bakx N, Hagelaar E, van der Sangen M, Theuws J, Hurkmans C. Artificial intelligence based treatment planning of radiotherapy for locally advanced breast cancer. Phys Imaging Radiat Oncol. 2021 Dec 1;20:111–116. Doi:10.1016/j.phro.2021.11.007. PMID: 34917779; PMCID: PMC8645926.
15. Jeong, C., Y. Goh, and J. Kwak, Challenges and opportunities to integrate artificial intelligence in radiation oncology: a narrative review. emj, 2024. 47(4): p. e49–0.DOI: https://doi.org/10.12771/emj.2024.e49
16. Poortmans PMP, Takanen S, Marta GN, Meattini I, Kaidar-Person O. Winter is over: The use of Artificial Intelligence to individualise radiation therapy for breast cancer. Breast. 2020 Feb;49:194–200. doi:10.1016/j.breast.2019.11.011. Epub 2019 Nov 26. PMID: 31931265; PMCID: PMC7375562.
17. van de Sande D, Sharabiani M, Bluemink H, Kneepkens E, Bakx N, Hagelaar E, van der Sangen M, Theuws J, Hurkmans C. Artificial intelligence based treatment planning of radiotherapy for locally advanced breast cancer. Phys Imaging Radiat Oncol. 2021 Dec 1;20:111–116. doi: 10.1016/j.phro.2021.11.007. PMID: 34917779; PMCID: PMC8645926.
18. Doolan PJ, Charalambous S, Roussakis Y, Leczynski A, Peratikou M, Benjamin M, Ferentinos K, Strouthos I, Zamboglou C and Karagiannis E (2023) A clinical evaluation of the performance of five commercial artificial intelligence contouring systems for radiotherapy. Front. Oncol. 13:1213068. doi: 10.3389/fonc.2023.1213068
19. Mariko Kawamura, Takeshi Kamomae, Masahiro Yanagawa, Koji Kamagata, Shohei Fujita, Daiju Ueda, Yusuke Matsui, Yasutaka Fushimi, Tomoyuki Fujioka, Taiki Nozaki, Akira Yamada, Kenji Hirata, Rintaro Ito, Noriyuki Fujima, Fuminari Tatsugami, Takeshi Nakaura, Takahiro Tsuboyama, Shinji Naganawa, Revolutionizing radiation therapy: the role of AI in clinical practice, Journal of Radiation Research, Volume 65, Issue 1, January 2024, Pages 1–9, https://doi.org/10.1093/jrr/rrad090
20. Weerarathna IN, Kamble AR, Luharia A. Artificial Intelligence Applications for Biomedical Cancer Research: A Review. Cureus. 2023 Nov 5;15(11):e48307. doi:10.7759/cureus.48307. PMID: 38058345; PMCID: PMC10697339.
21. Jiang S, Xue Y, Li M, Yang C, Zhang D, Wang Q, Wang J, Chen J, You J, Yuan Z, Wang X, Zhang X and Wang W (2022) Artificial Intelligence-Based Automated Treatment Planning of Postmastectomy Volumetric Modulated Arc Radiotherapy. Front. Oncol. 12:871871. doi:10.3389/fonc.2022.871871
22. Baroudi H, Brock KK, Cao W, Chen X, Chung C, Court LE, El Basha MD, Farhat M, Gay S, Gronberg MP, Gupta AC, Hernandez S, Huang K, Jaffray DA, Lim R, Marquez B, Nealon K, Netherton TJ, Nguyen CM, Reber B, Rhee DJ, Salazar RM, Shanker MD, Sjogreen C, Woodland M, Yang J, Yu C, Zhao Y. Automated Contouring and Planning in Radiation Therapy: What Is ‘Clinically Acceptable’? Diagnostics (Basel). 2023 Feb 10;13(4):667. doi: https://doi.org/10.3390/diagnostics13040667. PMID:36832155; PMCID: PMC9955359.
23. Zehua Wang, Ruichong Lin, Yanchun Li, Jin Zeng, Yongjian Chen, Wenhao Ouyang, Han Li, Xueyan Jia, Zijia Lai, Yunfang Yu, Herui Yao, Weifeng Su, Deep learning-based multi-modal data integration enhancing breast cancer disease-free survival prediction, Precision Clinical Medicine, Volume 7, Issue 2, June 2024, pbae012, https://doi.org/10.1093/pcmedi/pbae012
24. Gao, Y., Ventura-Diaz, S., Wang, X. et al. An explainable longitudinal multi-modal fusion model for predicting neoadjuvant therapy response in women with breast cancer. Nat Commun 15, 9613 (2024). https://doi.org/10.1038/s41467-024-53450-8
25. Shieh Y, Hu D, Ma L, Huntsman S, Gard CC, Leung JW, Tice JA, Vachon CM, Cummings SR, Kerlikowske K, Ziv E. Breast cancer risk prediction using a clinical risk model and polygenic risk score. Breast Cancer Res Treat. 2016 Oct;159(3):513–25. doi: 10.1007/s10549–016–3953–2. Epub 2016 Aug 26. PMID: 27565998; PMCID: PMC5033764.
26. Kara N. Maxwell et al.,Toward Application of Polygenic Risk Scores to Both Enhance and Deintensify Breast Cancer Screening. JCO 42, 1462–1465(2024). DOI:10.1200/JCO.24.00029
27. Lanchbury, J.S., Pederson, H.J. An apparent quandary: adoption of polygenics and gene panels for personalised breast cancer risk stratification. BJC Rep 1, 15 (2023). https://doi.org/10.1038/s44276-023-00014-w
28. Jiang, L., Xu, C., Bai, Y. et al. Autosurv: interpretable deep learning framework for cancer survival analysis incorporating clinical and multi-omics data. npj Precis. Onc. 8, 4 (2024). https://doi.org/10.1038/s41698-023-00494-6
29. Ting Jin, Nam D Nguyen, Flaminia Talos, Daifeng Wang, ECMarker: interpretable machine learning model identifies gene expression biomarkers predicting clinical outcomes and reveals molecular mechanisms of human disease in early stages, Bioinformatics, Volume 37, Issue 8, March 2021, Pages 1115–1124, https://doi.org/10.1093/bioinformatics/btaa935
30. Elisha Hughes et al., Integrating Clinical and Polygenic Factors to Predict Breast Cancer Risk in Women Undergoing Genetic Testing. JCO Precis Oncol 5, 307–316(2021). DOI:10.1200/PO.20.00246
31. Amber N Hurson, Parichoy Pal Choudhury, Chi Gao, Anika Hüsing, Mikael Eriksson, Min Shi, Michael E Jones, D Gareth R Evans, Roger L Milne, Mia M Gaudet, Celine M Vachon, Daniel I Chasman, Douglas F Easton, Marjanka K Schmidt, Peter Kraft, Montserrat Garcia-Closas, Nilanjan Chatterjee, for the B-CAST Risk Modelling Group, Prospective evaluation of a breast-cancer risk model integrating classical risk factors and polygenic risk in 15 cohorts from six countries, International Journal of Epidemiology, Volume 50, Issue 6, December 2021, Pages 1897–1911, https://doi.org/10.1093/ije/dyab036
32. Gurmessa DK, Jimma W. Explainable machine learning for breast cancer diagnosis from mammography and ultrasound images: a systematic review. BMJ Health Care Inform. 2024 Feb 2;31(1):e100954. doi: 10.1136/bmjhci-2023–100954. PMID: 38307616; PMCID: PMC10840064.
33. Eisemann, N., Bunk, S., Mukama, T. et al. Nationwide real-world implementation of AI for cancer detection in population-based mammography screening. Nat Med (2025). https://doi.org/10.1038/s41591-024-03408-6
34. Roberts E, Howell S, Evans DG. Polygenic risk scores and breast cancer risk prediction. Breast. 2023 Feb;67:71–77. doi: 10.1016/j.breast.2023.01.003. Epub 2023 Jan 10. PMID: 36646003; PMCID: PMC9982311.
35. Lee, A., Mavaddat, N., Wilcox, A.N. et al. BOADICEA: a comprehensive breast cancer risk prediction model incorporating genetic and nongenetic risk factors. Genet Med 21, 1708–1718 (2019). https://doi.org/10.1038/s41436-018-0406-9