pandani.ai is a multidisciplinary team of technologists, clinician–scientists and healthcare professionals.
Our AI-driven diagnostic enrichment tools augment the expertise of clinicians and enhance organisational workflows.
We believe expertly crafted diagnostic tools enrich clinical experience, improve patient outcomes, and elevate the quality of care.
We place people at the heart of our approach to AI technology.
Our goal is to empower clinicians with automated tools that provide accurate insights and reduce workflow burdens. Our technologies are designed to support the pivotal role of the clinician in patient care.
AI diagnostics should be fast, simple to use, transparent in function, and immediately accessible. We are committed to integrating flexible solutions that seamlessly adapt to health system contexts.
We build our tools for impact, delivering precision, greater workplace satisfaction, and streamlined operations for clinicians and healthcare providers.
Glaucoma is the leading cause of irreversible blindness worldwide. Although blindness from glaucoma can be prevented by timely intervention, it can be difficult to predict who will develop disease, and when diagnosed how quickly it will progress.
Our precision screening models can diagnose manifest glaucoma, project visual fields from retinal images (to readily identify structural and functional mismatches - important clinical "red flags"), and can predict the likelihood of disease progression. Our current research focus is the development of ancillary models which can readily identify other ocular and systemic conditions from retinal profiles to augment ocular care.
Published by our founding research team
This study developed and validated a generalised deep-learning computer vision model for glaucoma screening using over 18,000 retinal images. The best-performing model achieved excellent discrimination between glaucomatous and healthy optic discs (AUROC 0.992), with high sensitivity and specificity, and maintained good performance on an external validation dataset.
Although accuracy declined modestly on unseen data, likely due to dataset heterogeneity and inconsistent labelling, these results highlights both the promise of scalable AI-based screening and the need for validation on large, diverse datasets.
Published in Eye, November 2024
This study evaluated whether deep convolutional Generative Adversarial Networks can generate realistic synthetic optic disc images to improve deep-learning detection of glaucomatous optic neuropathy. Models trained solely on synthetic images achieved near-perfect internal performance but generalized poorly to real clinical data, indicating overfitting to synthetic features. In contrast, training on a mixed dataset of synthetic and real fundus images substantially improved external validation performance (AUROC ~86%), comparable to models trained only on real data.
Overall, these results suggest synthetic images are most valuable when used to augment—rather than replace—real clinical data in glaucoma detection models.
Published in Translational Vision Science & Technology, June 2024
This study developed a deep learning-based computer vision system to automatically assess fundus image quality and then precisely estimate the optic cup-to-disc ratio for glaucoma screening, without relying on optic disc segmentation. Trained on over 180,000 images, the best-performing Convolutional Neural Network achieved high accuracy for image gradability (97%) and strong cup-to-disc ratio regression performance (R² ≈ 0.85). When applied to external multiethnic datasets, the model demonstrated good glaucoma screening performance (AUROC ~0.87) with rapid, fully automated analysis.
These results demonstrate direct deep-learning-based cup-to-disc ratio estimation is efficient and reliable, though optimal cup-to-disc ratio thresholds for glaucoma detection vary across populations and clinical contexts.
Published in Ophthalmology Science, April 2024
This systematic review and meta-analysis evaluated the diagnostic accuracy of AI algorithms for glaucoma detection using fundus photographs and OCT images. Overall, AI demonstrated excellent performance, with a pooled AUROC of 96.3%, sensitivity of 92%, and specificity of 94%, showing comparable accuracy for fundus and OCT modalities. However, performance dropped substantially when models were tested on external datasets, highlighting limited generalisability and overestimation in internally validated studies.
This work highlights the fact that while AI has strong potential for glaucoma screening and clinical practice, standardised diagnostic protocols, robust external validation, and evaluation across diverse populations are essential before deployment.
Published in Journal of Glaucoma, May 2022
This review provides a comprehensive overview of how artificial intelligence—particularly machine learning and deep learning—is being applied across major ophthalmic diseases including glaucoma, diabetic retinopathy, age-related macular degeneration, and corneal ectasia.
It highlights that AI systems, especially convolutional neural networks, now achieve diagnostic performance comparable to expert clinicians in image-based screening and grading tasks, particularly for fundus photography and OCT imaging. It emphasises the potential of AI to improve efficiency, enable large-scale screening, and support clinical decision-making, especially in resource-limited settings.
Published in Clinical and Experimental Ophthalmology, August 2018
Digital pathology has significantly advanced cancer diagnosis by enabling high-resolution visualisation and assessment of tissue specimens. However, the manual analysis of these images remains labour-intensive and susceptible to observer variability, potentially resulting in inconsistencies in diagnosis and treatment decisions.
Our models provide anatomical pathologists with consistent, automated diagnostic capabilities. Our current research focus is the development of multimodal learning technologies that can provide prognostic insights from routine histological specimens – and in doing so, open an integrated, accessible computational pathway from histology to precision oncology.
Published by our founding research team
Accurate molecular profiling and prognostication from routine histopathology slides could transform precision oncology. We are developing an AI framework for combined predictions of 32 solid tumour types, TP53 biomarker detection, and survival prediction directly from Whole Slide Images (WSIs).
Preprint on medRxiv, November 2025
Quantification of the Ki-67 proliferation index is essential for breast cancer prognosis and treatment planning. Since H&E staining is routinely performed, accurately estimating Ki-67 from these slides could save resources by eliminating the need for additional immunohistochemical staining.
Further, the cross-modality model potentially quantifies molecular expression from morphological features using H&E-stained WSIs.
Published in Breast Cancer Research, November 2025
Skin cancer is a leading cause of preventable death in Australia, and various countries globally. Our model classifies the most common subtypes of skin cancer directly from WSIs. For non-melanoma cancer, our model can be deployed in regions with limited access to experienced dermatopathologists and a high incidence of the disease, particularly in low-resource settings.
Published in Computers in Biology and Medicine, September 2025
Prostate cancer is the second most frequent carcinoma among men and a leading cause of morbidity and mortality. We have developed a precise, objective, and reproducible model for identifying and grading malignant prostrate tissue.
Published in Prostate Cancer, March 2025
pandani.ai
Pandani Solutions Pty Ltd
ACN 681 297 825