This study investigates advanced deep learning methods to improve the detection of periportal fibrosis (PPF) in medical imaging. Schistosoma mansoni infection affects over 54 million individuals globally, predominantly in sub-Saharan Africa, with around 20 million experiencing chronic complications. PPF, present in up to 42% of these cases, is a leading outcome of chronic liver disease, significantly contributing to morbidity and mortality. Early and accurate detection is critical for timely intervention, yet conventional ultrasound diagnosis remains highly operator-dependent. We adapted and trained a convolutional neural network (CNN) using ultrasound images to automatically identify and classify PPF severity. The proposed approach achieved a diagnostic accuracy of 80%. Sensitivity and specificity reached 84% and 76%, respectively, demonstrating robust generalisability across varying image qualities and acquisition settings. These findings highlight the potential of deep learning to reduce diagnostic subjectivity and support scalable screening programmes. Future work will focus on validation with larger datasets and multi-class fibrosis grading to enhance clinical utility.

Background: While recent development of Artificial Intelligence (AI)‐based approaches have demonstrated to be effective in predicting risk of ADRD, these have mostly focused on AD subtype, aged and homogenous populations (Grueso et al, 2022; Rahim et al., 2023), thereby limiting their applicability to other types of ADRD and younger populations. Inspired by earlier work (Danso et al 2021), we propose an AI‐based deep‐learning framework for early detection of ADRD based on heterogeneous and diverse population from midlife (Figure 1). Method: We obtained two datasets from the European Prevention of Alzheimer's Dementia‐ EPAD ( n = 2096) and PREVENT Dementia Programme ( n = 700) available online (AD workbench, 2020). Following procedures described in Danso et al (2018) a harmonised cohort was curated containing individuals with no diagnosis of dementia. We then created three risk groups (High risk = ApoE4 allele and family history of AD; Medium risk = ApoE4 allele but no family history of AD; Low risk = no ApoE4 allele and no family history of AD) following the risk definition by Ritchie & Ritchie (2012). Convolutional Neural Network (CNN) and Long‐ Short Term Memory (LSTM) models were developed using 5‐fold cross validation and then applied optimisation procedures to obtain optimal parameters for the trained models. Result: The harmonisation resulted in a cohort ( n = 2796; mean age =62; range = 40 – 89years; female =57.5%, Caucasian = 95%), containing medical history, physiological, lifestyle, neuroimaging, and sociodemographic features. Overall, CNN outperformed LSTM by 7% points for accuracy and f1‐score (Table 1), with mean AUROC scores of 97% and 94% respectively (Figure 2), and mean validation loss scores (CNN = 0.36; LSTM =0.46). Conclusion: The superior performance of CNN is consistent with the literature and the relatively low validation loss demonstrates its generalisability. While this model is currently optimised for AD with limited features, a Transfer Learning paradigm is being employed to further train the CNN model to predict risk of other AD sub‐types after including BioHermes dataset into pipeline. Future work will also explore modifications of the CNN architecture for multimodal features with explainability capabilities.

This study investigates advanced deep learning methods to improve the detection 1 of periportal fibrosis (PPF) in medical imaging. Schistosoma mansoni infection affects over 2 54 million individuals globally, predominantly in sub-Saharan Africa, with around 20 3 million experiencing chronic complications. PPF, present in up to 42% of these cases, is 4 a leading outcome of chronic liver disease, significantly contributing to morbidity and 5 mortality. Early and accurate detection is critical for timely intervention, yet conventional 6 ultrasound diagnosis remains highly operator-dependent. We developed a convolutional 7 neural network (CNN) model trained on non-invasive ultrasound images to automatically 8 identify and classify PPF severity. The proposed approach achieved a diagnostic accuracy 9 of 80%. Sensitivity and specificity reached 80% and 84%, respectively, demonstrating robust 10 generalisability across varying image qualities and acquisition settings. These findings 11 highlight the potential of deep learning to reduce diagnostic subjectivity and support 12 scalable screening programs. Future work will focus on validation with larger datasets and 13 multi-class fibrosis grading to enhance clinical utility.