Deep learning (DL) is a subfield of artificial intelligence (AI) focused on developing algorithms that learn from data to perform some tasks that can aid humans in their daily life or work assignments. Research demonstrates the potential of DL in supporting pathologists with routine tasks like detecting breast cancer metastases and grading prostate cancer. However, a widespread adoption of DL technology in pathology labs has been slow for several reasons. DL models often exhibit performance variations across medical centres, patient subgroups, and even within the same centre over time. While collecting more data and retraining the algorithms seems like a straightforward solution, it is a costly and time-consuming process. Moreover, retraining DL systems with regulatory approvals is complex due to existing regulations. Another limitation of DL models is their inability to provide confidence estimates for predictions, leaving users in the dark about their reliability. Finally, establishing a close collaboration between the research community, vendors, and pathology labs is crucial for producing effective DL systems for patient care. However, this collaboration faces challenges like miscommunication, misalignment of goals, and misunderstanding priorities.

This thesis presents various approaches that could tackle the generalisation and reliability challenges faced by diagnostic DL systems for digital pathology with a strong emphasis on the clinical needs. To address the generalisation issues, an unsupervised approach to quantify expected changes in a model’s performance between two datasets is proposed. This approach can serve as an initial validation step before deploying diagnostic DL systems in clinical practice, reducing annotation costs. Additionally, an unsupervised framework based on generative models is proposed to identify substantially different inputs, known as out-of-distribution (OOD) samples. Detecting OOD samples plays a crucial role in enhancing the reliability of DL algorithms. Furthermore, several studies are conducted to explore what benefits uncertainty estimation could bring. Firstly, various uncertainty estimation approaches are extensively evaluated, focusing on identifying incorrect predictions and generalisability issues between medical centres and specific patient groups. In addition, the results reveal that combining uncertainty estimation methods with DL outputs leads to a more robust classification score, enhancing the overall performance and reliability of the classification process. Another study demonstrates that spatial uncertainty aggregation improves the effectiveness of uncertainty estimation in tumour segmentation tasks. This is evaluated on the detection of false negatives which may reduce the risk of missing tumour cells. Finally, the clinical prerequisites for developing and validating diagnostic DL systems for digital pathology are discussed, along with an overview of explainable AI techniques.

In conclusion, multiple approaches to facilitate the adoption of DL systems in clinical practice, addressing reliability, generalisability, and clinical needs aspects are discussed in this thesis. I believe that the extensive efforts in the research community will have a positive impact on the development, validation, and deployment of DL systems in digital pathology labs, empowering pathologists with trustworthy AI tools.

Breast cancer is the most common cancer among women in Sweden, and pathology is central to diagnosis and treatment planning. Advances in digital pathology, using high-resolution whole-slide images, (WSI), have  enabled the use of AI-based (artificial intelligence) image analysis tools that improves diagnostic efficiency and reproducibility. However, responsible implementation requires careful attention to clinical accountability, data governance and human oversight.

This thesis presents a multimodal evaluation framework and contributes to technical and medical insights to the field. In the first study a research database was constructed to support generalizable AI development, compromising six annotated imaging collections, comprising 754 WSIs and 24,043 pathology annotations, 110 radiology cases and 397 lesion annotations.

One study evaluated a human-in-the-loop (HITL) workflow, when pathologist assess cellproliferation as expressed by Ki-67 from an AI result. Even though AI showed reduced cell-level performance on local data, (F1 score 0.68 compared to 0.83), status agreement for visual estimation of Ki-67 performed significantly worse (Cohen’s κ 0.62) than digital image analysis (κ 0.84)  and HITL (κ 0.76). HITL reduced variability of the Ki-67 error and mitigated key limitations such as tumour heterogeneity, misidentification, and staining variability, while highlighting risks from user handling errors.

Subsequent studies introduced Feature Enhancing Zoom (FEZ), a visualization technique that amplifies stain patterns at low magnification. In a study with eight pathologists, FEZ improved task efficiency by 15% without compromising accuracy. A usability study with 16 pathologists confirmed high ratings, especially for stains requiring small object identification.

Finally, a methodology was evaluated to mitigate domain restraints during clinical implementation of a pretrained AI model for detecting metastases in lymph nodes. A locally curated dataset of 396 cases (4,462 WSIs) was used, with slides labelled by surgical procedure and lesion presence. Results showed that surgical procedure affects model performance, and retraining significantly improved generalization and thus reducing false positive predictions.

In summary, this thesis demonstrates how AI and digital pathology can be integrated into clinical workflows to enhance diagnostic precision.