Open research questions in AI in cancer detection

What the literature leaves open

• The development of models that can capture fine-grained histopathological details is a significant challenge. Prior work has focused on the development of various neural network architectures, but there is a need for more accurate and efficient models. The existing models are not optimized for high-resolution histopathological images and do not effectively capture cancer-relevant structures.

  Novel application of attention-enhanced hybrid efficient net for multi-scale histopathological cancer classification · 2026 · DOI

• The gap is due to the complexity of HSG images and the need for accurate and reliable classification. The paper also identifies the challenge of class imbalance in the dataset.

  YOLO11-based deep learning system for automated tubal patency classification in hysterosalpingography: a comparative study for clinical decision support · 2026 · DOI

• In particular, future prospective studies should focus on addressing the limitations identified in our work, such as the quality and standardization of EUS reports. This would ensure the systematic collection of all critical ultrasonographic variables, including intralesional vascularity and lesion stiffness.

  Explainable Machine Learning for Detecting Pancreatic Cancer from Structured Endoscopic Ultrasound Data: A Retrospective Multicenter Observational Study · 2026 · DOI

• The achievements of AI integration in stem cell histopathology need to be consciously oriented toward the need to fill existing gaps in data, experience, regulation, and translation. One of the main suggestions is investment in annotated, high-quality datasets. Currently, inconsistencies in staining procedures, imaging systems, and annotation habits negatively affect the applicability of AI models.

  Artificial Intelligence-Powered Histopathology in Stem Cell Research: Bridging Morphology, Function, and Omics · 2026 · DOI

• Federated learning provides a route for collaborative model development across institutions without centralizing raw whole-slide images, which may be useful when privacy and data governance requirements restrict direct data sharing.

  Artificial Intelligence-Powered Histopathology in Stem Cell Research: Bridging Morphology, Function, and Omics · 2026 · DOI

• Third, the cohort is single-source and small, and — most importantly — only 23 independent test patients underlie the reported intervals, so the effective evidence is limited regardless of the number of triplets.

  Integrating Temporal Reasoning in Vision-Language Models for Synthetic Pseudo-temporal Histopathology Sequence Classification on Breast Cancer Tissue Images · 2026 · DOI

• Future work will prioritize evaluation of a frozen MSF-Swin-ICFNet model on independent multicentre mammography cohorts, including heterogeneous acquisition systems and patient populations.

  MSF-Swin-ICFNet: An Image-Derived Multi-Scale Swin Transformer Framework with Cross-Scale Feature Fusion for Breast Lesion Classification and Segmentation in Mammograms · 2026 · DOI

• Further multi-institutional studies involving larger observer cohorts, including an independent external validation cohort, and prospective clinical implementation are warranted to better define the long-term clinical and educational impact of AI-assisted biomarker evaluation in breast cancer pathology.

  AI-Assisted Scoring Improves Interobserver Agreement in Breast Cancer Biomarker Evaluation · 2026 · DOI

• Whether the model generalizes to different countries, training systems, ultrasound protocols, or disease prevalence patterns remains to be established and requires international validation. Future prospective studies with randomized crossover designs and independent case sets are warranted to provide stronger evidence for the genuine incremental value of AI assistance.

  A CNN-based liver ultrasound plane recognition system with multimodal large language model-generated interpretable feedback · 2026 · DOI

• Staining and scanner variability—H&E color and scanner characteristics differ across laboratories, and although color augmentation mitigates this, robustness to domain shift across institutions has not been quantified. (11) Explainability not yet formally validated—Grad-CAM maps were assessed qualitatively and have not been reviewed by pathologists or compared against region annotations.

  Adaptive Multi-Scale Feature Fusion with Hybrid Representation Learning to Classify and Retrieve Histopathological Images · 2026 · DOI

• Accordingly, the repeatability of S-Detect across operators with different levels of experience remains unclear. Third, the study was limited to the analysis of identical stored static images rather than real-time ultrasound examinations.

  Intrasystem Repeatability of S-Detect for Breast Ultrasound Classification With Identical Static Images: Single-Center Retrospective Repeatability Study · 2026 · DOI

• One area for improvement is addressing a key limitation in the proposed methodology related to potential information leakage during the training of hybrid models, which poses a critical risk to the model’s reliability and practical utility. Overall, the RF and SVM hybrids achieved computationally satisfactory levels of accuracy; however, these levels remain insufficient from a clinical standpoint.

  Exploratory Study on Hybrid Systems Performance: A First Approach to Hybrid ML Models in Breast Cancer Classification · 2026 · DOI

• One major direction for future work is the incorporation of reinforcement learning. By leveraging reinforcement learning, the model can iteratively learn from its own predictions and feedback, refining its decision-making process over time.

  AI Method for Classification of Diagnosis of Near-Infrared Breast Lesion Images · 2026 · DOI

• Future work will investigate improving the fidelity of the learned distributions through alternative flow-matching formulations, interpolation paths, and ODE solvers, as well as incorporating formal privacy guarantees that would complement the empirical evidence of non-memorization observed in this study.

  Flow Matching for Generating Weakly Labeled Bags of Foundation-Model Mammography Representations · 2026 · DOI

• Data augmentation and synthesis Enhancing training data via augmentation (morphing cell images, simulating artifacts) or using generative models (e.g., GANs to create synthetic cytology images) could address data scarcity and make models more robust to variations, thus reducing uncertainty in real-world applications.

  Uncertainty-aware and interpretable deep learning in cytopathology: toward trustworthy AI-powered diagnostics · 2026 · DOI

• Several important limitations must be acknowledged before any clinical interpretation of these results. First, all experiments were conducted on a single publicly available dataset; the model has not been validated on independent, multi-institutional cohorts, and its generalisation to slides from different scanners, staining protocols, or patient populations remains unestablished.

  Differential sparse attention framework for accurate ovarian cancer subtype classification using DenseNet121 · 2026 · DOI

• In contrast, despite US is widely used in gynecological practice, its potential for LVSI prediction has been largely underexplored. Although the higher training AUC suggests some degree of model optimism, further validation in larger multicenter cohorts is warranted.

  Peritumoral ultrasound-driven multimodal AI model for predicting lymphovascular space invasion and prognosis in cervical cancer · 2026 · DOI

• The generalizability, clinical accuracy, and practical value of the pipeline remain to be established through independent, prospective, and multi-center validation. Additionally, the ability to train using only nucleus-level annotations, without requiring slide-level labels, enhances its practicality in real-world settings where such annotations are scarce.

  Automated whole slide Allred score estimation for estrogen receptor assessment in breast cancer · 2026 · DOI

• Limitations and Future Validation A key limitation of the present study is that the original NPC dataset was randomly partitioned at the patch level rather than at the patient or WSI level.

  DSR-VGG: A Lightweight Depthwise Separable Residual VGG Architecture for Nasopharyngeal Histopathological Image Classification · 2026 · DOI

• Our findings are consistent with the limited literature on CAD repeatability in mammography and extend those observations to ultrasound, a modality with inherently greater acquisition variability, by showing that, under standardized acquisition and a controlled analysis environment, a commercial integrated CAD module can be highly stable.

  Intrasystem Repeatability of S-Detect for Breast Ultrasound Classification With Identical Static Images: Single-Center Retrospective Repeatability Study · 2026 · DOI

• This highlights the potential of AI-assisted pathology not only for automating feature extraction but also for uncovering biologically and clinically relevant tissue structures that may be under-recognized in routine manual assessment.

  Multiview deep-learning-enabled histopathology for prognostic and therapeutic stratification in stage II colorectal cancer: A retrospective multicenter study · 2026 · DOI

• Further research can be done to improve the model's performance on other types of cancer, - The model can be applied to other medical imaging tasks, - The attention mechanism can be further optimized for other types…

  Novel application of attention-enhanced hybrid efficient net for multi-scale histopathological cancer classification · 2026 · DOI

• Exploring other potential applications of the model. Developing more efficient and accurate models for tumor segmentation and subtype classification. Investigating the use of other deep learning architectures for RCC diagnosis.

  Deep learning-based segmentation and subtype prediction of renal cell carcinoma on contrast-enhanced CT · 2026 · DOI

• Conventional quantification of immune cells in tissue specimens has relied on manual counting under the microscope or simple thresholding of digitized images, - Whole-image AI approaches capture global textural patterns but lack cellular identity and spatial resolution, - Cu-Cyto operates on standard bright-field immunohistochemistry rather than a dedicated acquisition…

  From cell counts to cellular interactions: Cu-Cyto and the co-localization index as a spatial framework for the tumor immune microenvironment of rectal cancer · 2026 · DOI

• Further development of Cu-Cyto and CLI for routine clinical deployment, - Investigation of the Co-Localization Index in other types of cancer, - Exploration of the potential of Cu-Cyto and CLI in patient stratification…

  From cell counts to cellular interactions: Cu-Cyto and the co-localization index as a spatial framework for the tumor immune microenvironment of rectal cancer · 2026 · DOI

Most-cited papers in AI in cancer detection

• Diagnostic Assessment of Deep Learning Algorithms for Detection of Lymph Node Metastases in Women With Breast Cancer · JAMA · 2017 · 2,894 citations
• Segment anything in medical images · Nature Communications · 2024 · 2,440 citations
• Data-efficient and weakly supervised computational pathology on whole-slide images · Nature Biomedical Engineering · 2021 · 2,234 citations
• Towards a general-purpose foundation model for computational pathology · Nature Medicine · 2024 · 1,185 citations
• Deep learning in histopathology: the path to the clinic · Nature Medicine · 2021 · 874 citations
• Deep learning in cancer diagnosis, prognosis and treatment selection · Genome Medicine · 2021 · 867 citations
• Medical image segmentation using deep learning: A survey · IET Image Processing · 2022 · 791 citations
• Automated deep-learning system for Gleason grading of prostate cancer using biopsies: a diagnostic study · The Lancet Oncology · 2020 · 680 citations
• Pathomic Fusion: An Integrated Framework for Fusing Histopathology and Genomic Features for Cancer Diagnosis and Prognosis · IEEE Transactions on Medical Imaging · 2020 · 662 citations
• A whole-slide foundation model for digital pathology from real-world data · Nature · 2024 · 656 citations

Most recent work

• Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial · The Lancet · 2026
• CellViT++: Energy-efficient and adaptive cell segmentation and classification using foundation models · Computer Methods and Programs in Biomedicine · 2026
• Explainability‐Aligned Reliability‐Weighted Fuzzy Ensemble for Automated Cervical Cancer Classification · International Journal of Intelligent Systems · 2026
• Confounding factors and biases abound when predicting molecular biomarkers from histological images · Nature Biomedical Engineering · 2026
• Diagnostic accuracy, fairness and clinical implementation of AI for breast cancer screening: results of multicenter retrospective and prospective technical feasibility studies · Nature Cancer · 2026
• Deep learning on histopathological images to predict breast cancer recurrence risk and chemotherapy benefit: a multicentre, model development and validation study · The Lancet Oncology · 2026
• ms-mamba: Multi-scale mamba for time-series forecasting · Neurocomputing · 2026
• Artificial intelligence and multi-omics convergence in breast cancer: Revolutionizing diagnosis, prognostication, and precision oncology · Critical Reviews in Oncology/Hematology · 2026
• An agentic framework for autonomous scientific discovery in cancer pathology · Nature Medicine · 2026
• Knowledge-enhanced pretraining for vision-language pathology foundation model on cancer diagnosis · Cancer Cell · 2026

Related topics in Computer Science