Preparing Data for Artificial Intelligence in Pathology with Clinical-Grade Performance

Enhancing Medical Imaging Segmentation with GB-SAM: A Novel Approach to Tissue Segmentation Using Granular Box Prompts

Recent advances in foundation models have revolutionized model development in digital pathology, reducing dependence on extensive manual annotations required by traditional methods. The ability of foundation models to generalize well with few-shot learning addresses critical barriers in adapting models to diverse medical imaging tasks. This work presents the Granular Box Prompt Segment Anything Model (GB-SAM), an improved version of the Segment Anything Model (SAM) fine-tuned using granular box prompts with limited training data. The GB-SAM aims to reduce the dependency on expert pathologist annotators by enhancing the efficiency of the automated annotation process. Granular box prompts are small box regions derived from ground truth masks, conceived to replace the conventional approach of using a single large box covering the entire H&E-stained image patch. This method allows a localized and detailed analysis of gland morphology, enhancing the segmentation accuracy of individual glands and reducing the ambiguity that larger boxes might introduce in morphologically complex regions. We compared the performance of our GB-SAM model against U-Net trained on different sizes of the CRAG dataset. We evaluated the models across histopathological datasets, including CRAG, GlaS, and Camelyon16. GB-SAM consistently outperformed U-Net, with reduced training data, showing less segmentation performance degradation. Specifically, on the CRAG dataset, GB-SAM achieved a Dice coefficient of 0.885 compared to U-Net's 0.857 when trained on 25% of the data. Additionally, GB-SAM demonstrated segmentation stability on the CRAG testing dataset and superior generalization across unseen datasets, including challenging lymph node segmentation in Camelyon16, which achieved a Dice coefficient of 0.740 versus U-Net's 0.491. Furthermore, compared to SAM-Path and Med-SAM, GB-SAM showed competitive performance. GB-SAM achieved a Dice score of 0.900 on the CRAG dataset, while SAM-Path achieved 0.884. On the GlaS dataset, Med-SAM reported a Dice score of 0.956, whereas GB-SAM achieved 0.885 with significantly less training data. These results highlight GB-SAM's advanced segmentation capabilities and reduced dependency on large datasets, indicating its potential for practical deployment in digital pathology, particularly in settings with limited annotated datasets.

Detailed Image Data Quality and Cleaning Practices for Artificial Intelligence Tools for Breast Cancer

Standardizing image-data preparation practices to improve accuracy/consistency of AI diagnostic tools.

Algorithm-assisted diagnosis of Hirschsprung's disease - evaluation of robustness and comparative image analysis on data from various labs and slide scanners

BACKGROUND

Differences in the preparation, staining and scanning of digital pathology slides create significant pre-analytic variability. Algorithm-assisted tools must be able to contend with this variability in order to be applicable in clinical practice. In a previous study, a decision support algorithm was developed to assist in the diagnosis of Hirschsprung's disease. In the current study, we tested the robustness of this algorithm while assessing for pre-analytic factors which may affect its performance.

METHODS

The decision support algorithm was used on digital pathology slides obtained from four different medical centers (A-D) and scanned by three different scanner models (by Philips, Hamamatsu and 3DHISTECH). A total of 192 cases and 1782 slides were used in this study. RGB histograms were constructed to compare images from the various medical centers and scanner models and highlight the differences in color and contrast.

RESULTS

The algorithm was able to correctly identify ganglion cells in 99.2% of cases, from all medical centers (All scanned by the Philips slide scanner) as well as 95.5% and 100% of the slides scanned by the 3DHISTECH and Hamamatsu brand slide scanners, respectively. The total error rate for center D was lower than the other medical centers (3.9% vs 7.1%, 10.8% and 6% for centers A-C, respectively), the vast majority of errors being false positives (3.45% vs 0.45% false negatives). The other medical centers showed a higher rate of false negatives in relation to false positives (6.81% vs 0.29%, 9.8% vs 1.2% and 5.37% vs 0.63% for centers A-C, respectively). The total error rates for the Philips, Hamamatsu and 3DHISTECH brand scanners were 3.9%, 3.2% and 9.8%, respectively. RGB histograms demonstrated significant differences in pixel value distribution between the four medical centers, as well as between the 3DHISTECH brand scanner when compared to the Philips and Hamamatsu brand scanners.

CONCLUSIONS

The results reported in this paper suggest that the algorithm-based decision support system has sufficient robustness to be applicable for clinical practice. In addition, the novel method used in its development - Hierarchial-Contexual Analysis (HCA) may be applicable to the development of algorithm-assisted tools in other diseases, for which available datasets are limited. Validation of any given algorithm-assisted support system should nonetheless include data from as many medical centers and scanner models as possible.