The increased workload in pathology laboratories today means automated tools such as artificial intelligence models can be useful, helping pathologists with their tasks. In this paper, we propose a segmentation model (DRU-Net) that can provide a delineation of human non-small cell lung carcinomas and an augmentation method that can improve classification results. The proposed model is a fused combination of truncated pre-trained DenseNet201 and ResNet101V2 as a patch-wise classifier, followed by a lightweight U-Net as a refinement model. Two datasets (Norwegian Lung Cancer Biobank and Haukeland University Lung Cancer cohort) were used to develop the model. The DRU-Net model achieved an average of 0.91 Dice similarity coefficient. The proposed spatial augmentation method (multi-lens distortion) improved the Dice similarity coefficient from 0.88 to 0.91. Our findings show that selecting image patches that specifically include regions of interest leads to better results for the patch-wise classifier compared to other sampling methods. A qualitative analysis by pathology experts showed that the DRU-Net model was generally successful in tumor detection. Results in the test set showed some areas of false-positive and false-negative segmentation in the periphery, particularly in tumors with inflammatory and reactive changes. In summary, the presented DRU-Net model demonstrated the best performance on the segmentation task, and the proposed augmentation technique proved to improve the results.
Classification of lung cancer subtypes is a critical clinical step; however, relying solely on H&E-stained histopathology images can pose challenges, and additional immunohistochemical analysis is sometimes required for definitive subtyping. Digital pathology facilitates the use of artificial intelligence for automatic classification of digital tissue slides. Automatic classification of Whole Slide Images (WSIs) typically involves extracting features from patches obtained from them. The aim of this study was to develop a WSI classification framework utilizing an optimizable kernel to encode features from each patch from a WSI into a desirable and adjustable latent space using an evolutionary algorithm. The encoded data can then be used for classification and segmentation while being computationally more efficient. Our proposed framework is compared with a state-of-the-art model, Vim4Path, on an internal and external dataset of lung cancer WSIs. The proposed model outperforms Vim-S16 in accuracy and F1 score at both and magnification levels on the internal test set, with the highest accuracy (0.833) and F1 score (0.721) at . On the external test set, Vim-S16 at achieves the highest accuracy (0.732), whereas OKEN-DenseNet121 at has the best F1 score (0.772). In future work, finding a dynamic way to tune the output dimensions of the evolutionary algorithm would be of value.
Whole slide imaging has transformed the field of pathology by enabling high-resolution digitization of histopathological slides. However, the large image size and variability in morphology, tissue processing, and imaging can pose challenges for robust computational analysis. When working with specific tasks in digital pathology, conventional feature extractors pretrained on general images may not provide features as relevant as those trained on histopathological images. To address this, foundation models pretrained on histopathological images have been developed. Yet, their large size and computational demands might limit widespread adoptions to specific tasks. To facilitate the low-cost adoption of these models, we utilized low-rank adaptation for finetuning the model and developed evolving prototype-based multiple instance learning (EP-MIL). Our method’s capabilities were demonstrated by applying it to the classification of two histological subtypes of lung cancer. The results show that our approach achieves competitive performance when benchmarked against a state-of-the-art technique (CLAM), while offering improvements in efficiency. Specifically, our proposed method requires 8.3 times less training runtime compared with CLAM, uses less than 200.0 MB of memory during training, and enables 73.8 times faster inference runtime. These efficiency gains, combined with competitive performance, suggest that utilizing evolving prototypes with LoRA-tuned foundation models offers a more efficient and practical approach for broader use of foundation models in resource-constrained clinical settings.