International Journal of All Research Education & Scientific Methods

An ISO Certified Peer-Reviewed Journal

ISSN: 2455-6211

Latest News

Visitor Counter
12289121746

Deep Learning-Based Clinical Decision Support...

You Are Here :
> > > >
Deep Learning-Based Clinical Decision Support...

Deep Learning-Based Clinical Decision Support System for Hair and Scalp Disease Diagnosis

Author Name : Mohammad Uzair, Vansh Jaiswal, Pallavi Khadse, Mansi Purme, Pratham Ahuja, Dr. Ruchita Kale

ABSTRACT Scalp and hair disorders are difficult to diagnose because different conditions look alike, manual trichoscopy is subjective, and dermatologists are scarce outside major cities. This paper presents ScalpAI, a two-tier clinical decision-support system that couples deep-learning classification with computer-vision trichometry. The classification tier fine-tunes an EfficientNet-B0 network, built from compound-scaled mobile inverted bottleneck residual blocks, on 2,250 clinical and trichoscopic images covering fourteen scalp conditions and healthy controls. The trichometry tier combines contrast-limited adaptive histogram equalization, Sauvola thresholding, morphological filtering, skeletonization and distance transforms to measure hair count, shaft caliber in micrometers, follicular-unit grouping and an Anisotrichosis Index. Gradient-weighted class activation maps show which image regions drive each prediction. Over the 2,250-image cohort the classifier reached 84.84% Top-1 and 90.13% Top-3 accuracy with 98.92% macro specificity, and automated trichometry correlated with manual reticle measurements on 100 images (Pearson r from 0.89 to 0.94 across four metrics). The complete pipeline runs in about 452 ms on a laptop CPU and is delivered through a Django web portal that generates PDF reports. Because the reported metrics were not computed on a strictly held-out test set, they should be treated as preliminary.