Chandigarh: A joint study by Punjabi University, Patiala, and PGIMER, Chandigarh, has developed artificial intelligence-based methods to aid diagnosis of Autoimmune Blistering Diseases (AIBDs). These rare skin disorders are marked by severe blistering and are considered clinically complex and difficult to diagnose.
Study Background and Ethical Approval
The university officials said that the study received formal ethical clearance from the Institutional Ethics Committees of both institutions and approval from the Collaborative Research Committee of PGIMER. The research was carried out by PhD scholar Manbir Singh under the supervision of Maninder Singh from the department of computer science at Punjabi University, Patiala, and the co-supervision of Prof Dipankar De from the department of dermatology at PGIMER, Chandigarh.
Challenges in Diagnosing AIBDs
Manbir Singh said diagnosing AIBDs was exceptionally difficult because accurate confirmation required multiple specialised investigations. Many tests were highly expensive and time-consuming, leading to significant delays in confirming the disease. The diagnostic process became more complicated because AIBDs comprise several subtypes that share heavily overlapping clinical features.
Existing Diagnostic Methods
Maninder Singh explained that Direct Immunofluorescence is the gold standard test for confirming AIBDs, complemented by Indirect Immunofluorescence and Enzyme-Linked Immunosorbent Assay tests. However, these investigations are constrained by high costs, long turnaround times, and the need for high-level technical expertise, and are typically available only in specialised tertiary care institutions. He added that the new AI tool would be immensely helpful for clinicians in primary care settings and rural healthcare centres, where access to specialised laboratory tests and expert dermatology consultation is highly limited.
AI Research on Rare Skin Diseases
The research team noted that extensive AI research has focused on classifying common skin diseases, but AI research on rare and complex conditions such as AIBDs remains severely limited, primarily because of the absence of a clinically validated, publicly available dataset. To address this, the team created a clinically validated dataset by collecting authentic clinical images from patients diagnosed and treated at the dermatology department of PGIMER, Chandigarh.
Dataset Creation and Model Evaluation
Prof Dipankar De said the images were meticulously annotated under expert supervision only after complete clinical and diagnostic confirmation of the disease. He stated that the study evaluated the performance of classical machine learning, hybrid models, and advanced deep learning approaches, carrying out a comprehensive evaluation of nearly 240 model configurations. To assess practical utility, the diagnostic performance of the developed AI models was compared directly against dermatologists with varying levels of professional experience. The developed models consistently outperformed the participating dermatologists in accurately classifying distinct AIBD subtypes.
Future Plans
The research team said it aims to expand the dataset by incorporating more clinical images of rare AIBD subclasses and add comprehensive patient metadata, including age, sex, demographic backgrounds, laboratory test reports, and specific lesion locations. Future iterations will integrate Vision Transformer-based models and longitudinal disease monitoring systems to boost classification performance and real-world clinical applicability.
University Leadership's Remarks
The vice-chancellor of Punjabi University, Patiala, said such innovative research stands as a prime example of how modern technology can revolutionise human healthcare and medical science. He noted that the joint venture between Punjabi University and PGIMER will prove to be a true boon for the general public and underserved primary health sectors in the times to come.
