ML-based Histopathological Analysis of Stained Liver and GIT Tissues: A Pilot Implementation of Model Performance in a Sub-Saharan African Laboratory

Histopathological analysis is crucial in diagnosing liver and gastrointestinal (GIT) diseases, yet traditional methods can be time-intensive and subjective. The advancements in artificial intelligence, such as Convolutional Neural Networks (CNNs), offer innovative solutions to these challenges. The VGG16 model of CNN is a deep learning architecture known for its ability to extract detailed hierarchical features. This pilot study evaluates the effectiveness of CNNs in histopathology of liver and gastrointestinal tract (GIT) tissue diagnosis, focusing on their accuracy and usability in sub-Saharan African laboratories. Fourteen liver and fourteen GIT tissue samples were analysed using the Visual Geometry Group 16 (VGG16)-based CNN model trained and validated on photomicrography images from randomly selected tissue samples to classify cellular patterns and structures across three histopathology stains. Model performance analysis was done to evaluate performance across tissue types. The liver model achieved 100% training accuracy but had limited generalisation with 62.5% validation accuracy (95% CI: 52.5–71.6%), showing overfitting (37.5% gap). The GIT model performed better, with 96.43% training and 85.71% validation accuracy (95% CI: 77.4–91.5%). Statistical analysis confirmed significant performance (liver: p = 0.0062; GIT: p < 0.0001). While the GIT model is promising for clinical validation, the liver model requires refinement through regularisation techniques and diverse datasets. Implementation in developing regions faces challenges including limited computing resources. Future work should expand datasets and interdisciplinary collaboration to enhance clinical applicability. This need is especially critical in sub-Saharan Africa, where limited computing infrastructure and shortage of pathologists impact timely diagnosis.