An AI-Based Multi-Stage Approach for Androgenetic Alopecia Assessment from Low-Magnification Scalp Images

Mahmoud Raslan; Nada Omar; Omar Khaled; Tarek Waleed; Mohamed Hazem; Rania Mounir; Solwan Elsamanoudy; Ahmed Mourad

cs.CVOct 1, 2026

An AI-Based Multi-Stage Approach for Androgenetic Alopecia Assessment from Low-Magnification Scalp Images

Authors: Mahmoud Raslan, Nada Omar, Omar Khaled, Tarek Waleed, Mohamed Hazem, Rania Mounir, Solwan Elsamanoudy, Ahmed Mourad,

Organizations: Faculty of Engineering, Cairo University, Giza, Egypt · Dermatology Department, Faculty of Medicine, Cairo University, Cairo, Egypt

Abstract

Androgenetic alopecia (AGA) is characterized by patterned follicular miniaturization, increased single-hair follicular units, and altered hair-shaft diameter. We present an automated quantitative scalp-analysis and clinical decision-support framework combining FU localization, ordinal visible-shaft counting, calibrated shaft-width estimation, regional aggregation, and an interpretable rule layer. The clinical cohort comprised 243 patients (127 AGA, 116 non-AGA), while the computer-vision experiments used 160 expert-annotated patients, 2,400 trichoscopic images, and approximately 158,000 FU annotations. Under patientdisjoint evaluation, YOLOv8m achieved test [email protected]=0.920 and recall=0.860; EfficientNet-B5 with a support-map channel achieved 87.0% expert-box count accuracy (macro F1=0.85). A separate 500-image set was processed end-to-end with detector-generated boxes, yielding MAE of 6.56 for follicle detection and 16.59 for follicle classification relative to human-expert annotations. The system is intended to assist, rather than replace, dermatologist interpretation.

Figures & tables

Panel of Figure 1 — Fig. 1: Compact high-level workflow. Each stage passes interpretable measurements to the next module, ending with a clinician-facing rule-based summary.

Characteristic	AGA	Non-AGA
Patients	127	116
Female / male	102 / 25	104 / 12
Mean age (years)	29.15	26.58
Median age (years)	27	26
FPHL	102	0
MPHL	25	0

TABLE I: Clinical cohort demographics and diagnosis composition.

Statistic	Count
Expert-annotated imaging patients	160
Images per imaging patient	15
Total trichoscopic images	2,400
Anatomical landmarks	5
Images per landmark	480
Expert FU annotations	$\approx$ 158,000

TABLE II: Computer-vision dataset composition and derived task subsets.

Fig. 2: Representative trichoscopic image with expert follicular-unit bounding boxes.

Panel of Figure 5 — Fig. 3: Representative expert-annotated follicular-unit crops for the ordinal count task.

Fig. 4: Slicing-aided preprocessing: overlapping tiles increase the relative scale of small follicular units.

Output	Rule	Interpretation
Consistent with AGA	$C_{1}\land C_{2}$	Both criteria support AGA
Possible AGA	$C_{1}\oplus C_{2}$	One criterion positive
AGA not supported	$\neg C_{1}\land\neg C_{2}$	Neither criterion positive

TABLE III: Clinician-facing AGA support categories.

Model	Split	P	R	F1	[email protected]	[email protected]:0.95	Params
YOLOv8l-P2	Val	0.672	0.712	0.691	0.684	0.234	42.8M
YOLOv8l-P2	Test	0.681	0.707	0.694	0.694	0.233	–
YOLO11s-P2	Val	0.673	0.687	0.680	0.565	0.166	12.5M
YOLO11s-P2	Test	0.686	0.699	0.692	0.587	0.173	–
YOLO26s-P2	Val	0.560	0.577	0.568	0.486	0.143	9.4M
YOLO26s-P2	Test	0.570	0.580	0.575	0.495	0.143	–

TABLE IV: Initial YOLO-family follicular-unit screening results.

Model	Acc.	Macro P	Macro R	Macro F1	Val. Acc.
EfficientNet-B2	0.827	0.81	0.82	0.82	0.819
EfficientNet-B2+S	0.835	0.83	0.82	0.82	0.832
EfficientNet-B3	0.839	0.83	0.83	0.83	0.832
EfficientNet-B3+S	0.842	0.83	0.83	0.83	0.831
EfficientNet-B5	0.860	0.84	0.85	0.85	0.845
EfficientNet-B5+S	0.870	0.86	0.85	0.85	0.849

TABLE V: EfficientNet classifier comparison under expert-verified FU localization. “+S” denotes inclusion of the support map as a fourth input channel.

Method	Valid	ICC	Bias ( $\mu$ m)	Time (s/img)
Global branch-based	15/15	0.818	+0.8	142.8
Root-anchored support-map	158/160	0.905	+4.087	12.0

TABLE VI: Exploratory width-estimation results from the manual-reference experiment.

Fig. 5: Representative end-to-end output. FU boxes are color-coded by predicted ordinal count and measured shaft segments are rendered according to diameter category.

Medical Image Analysis

Medical image analysis uses computational methods to interpret clinical images, measure anatomy and disease, and support diagnosis and treatment.

See all 292 papers

CG-HAF: An Interpretable Global-Local Lesion-Burden Fusion Framework for Ordinal Acne Severity Grading in Agentic Skincare Support
Sep 25, 2026Muhammad Muhtasim Shahriar, Md. Naimur Asif Borno, Saad Aloteibi +1Medical Image Analysis Explainable Medical Image Analysis
Department of Computer Science, International Islamic University Chittagong (IIUC), Chittagong, Bangladesh · Department of Mechatronics, Rajshahi University of Engineering and Technology (RUET), Rajshahi, Bangladesh · Department of Computer Science and Engineering, College of Applied Studies, King Saud University, Riyadh, 11437, Saudi Arabia +3
Clinical Validation of the Melanoscope AI Mobile Dermoscopy Clinical Decision Support System
May 26, 2026Elena Sergeevna Kozachok, Sergey Sergeevich SereginMedical Image Classification Clinical Triage
1Ivannikov Institute for System Programming of the Russian Academy of Sciences (ISP RAS),2026 25 Alexander Solzhenitsyn str., Moscow, 109004, Russian Federation · 2Orel Regional Oncology Dispensary,May 2 Ippodromny lane, Orel, Russian Federation
Retinal OCTA Phenotyping with LLM Reporting for Alzheimer's Disease
Sep 4, 2026Progga Paromita Dutta, Jeba Maliha, Md Rafiul KabirExplainable Medical Image Analysis Optical Coherence Tomography
Department of Computer Science, Columbia University, New York, NY, USA · School of Engineering and Technology, Central Michigan University, Mount Pleasant, MI, USA

Sep 25, 2026J/K move · Enter open · S save

CG-HAF: An Interpretable Global-Local Lesion-Burden Fusion Framework for Ordinal Acne Severity Grading in Agentic Skincare Support

Muhammad Muhtasim Shahriar, Md. Naimur Asif Borno, Saad Aloteibi, Mohammad Ali Moni

Department of Computer Science, International Islamic University Chittagong (IIUC), Chittagong, Bangladesh · Department of Mechatronics, Rajshahi University of Engineering and Technology (RUET), Rajshahi, Bangladesh · Department of Computer Science and Engineering, College of Applied Studies, King Saud University, Riyadh, 11437, Saudi Arabia+3

Medical Image Analysis Explainable Medical Image Analysis Medical Diagnosis Feature Fusion Skin Lesion Classification Ordinal Regression

Open paper AskPDF arXiv

Ordinal acne severity grading requires distinguishing visually similar neighboring grades while jointly weighing holistic facial appearance and localized lesion burden - evidence that most existing approaches collapse into a single opaque representation. We introduce CG-HAF, a global-local fusion framework that instead keeps this evidence explicit: averaged holistic severity probabilities from independently trained classifiers are combined with structured lesion-burden descriptors from an object detector (lesion count, detection confidence, lesion area) into a compact representation, from which a lightweight, interpretable classifier produces the final grade. On a widely used benchmark, this fusion yields a clear, statistically supported improvement over global-evidence-only baselines, with the largest gains on the most severe cases. Testing on an independent dataset with a different grading standard shows that strong within-dataset performance does not transfer automatically, and a follow-up diagnostic attributes much of this gap to mismatched grading criteria rather than detection failure alone. These findings support interpretable global-local fusion as an effective strategy for ordinal acne grading while highlighting criterion alignment as key to cross-dataset portability, with a further illustration of how the resulting severity signal can support transparent, non-diagnostic decision-making in skincare applications.

Abstract

Medical Image Analysis

Explore similar work

CG-HAF: An Interpretable Global-Local Lesion-Burden Fusion Framework for Ordinal Acne Severity Grading in Agentic Skincare Support

Clinical Validation of the Melanoscope AI Mobile Dermoscopy Clinical Decision Support System

Retinal OCTA Phenotyping with LLM Reporting for Alzheimer's Disease