Smartphone-Enabled Depth-Aware Transillumination Imaging for Detection of Chicken Breast Fillet Myopathies in Chicken Fillets
Authors: Chaitanya Pallerla, Ramesh Bahadur Bist, Yihong Feng, Yang Tian, Shiva Paudel, Casey M. Owens, Cengiz Koparan, Jeyamkondan Subbiah, Dongyi Wang
Organizations: Department of Biological and Agricultural Engineering, University of Arkansas, Fayetteville, AR, USA, 72701 · Department of Food Science, University of Arkansas, Fayetteville, AR, USA, 72701 · Department of Poultry Science, University of Arkansas, Fayetteville, AR, USA, 72701
Woody Breast (WB) myopathy is a major quality defect in broiler chickens characterized by fibrosis, increased tissue hardness, and altered structural properties that affect product quality. Conventional palpation is subjective and labor-intensive, while many non-destructive methods require specialized equipment or substantial computational resources. This study developed a smartphone-based transillumination and 3D point-cloud framework for non-destructive WB severity classification. A total of 267 chicken breast fillets, including normal breast (NB, n = 87), mild WB (MWB, n = 85), and severe WB (SWB, n = 95), were evaluated using RAW transillumination images and 3D point clouds acquired with an iPhone 16 Pro. A total of 91 features, comprising 14 image-derived, 76 geometric, and one physics-informed Beer-Lambert attenuation feature, were extracted. Partial correlation controlling for thickness revealed a significant association between image characteristics and WB severity (r = 0.413, p < 0.0001), indicating optical information independent of fillet geometry. SWB tissue exhibited approximately 26% greater light attenuation per unit thickness than NB. The 2D-only models achieved approximately 82% test accuracy, while the multimodal ensemble combining Fusion-DGCNN, XGBoost, and physics-guided logistic regression (PGLR) achieved approximately 90% accuracy. These findings demonstrate the complementary value of optical, geometric, and physics-informed features and establish the feasibility of smartphone-based, non-destructive WB severity assessment.
Figures & tables
Fold
Train
Val
Test
Total
0
181 (NB 59 / MWB 58 / SWB 64)
32 (NB 10 / MWB 10 / SWB 12)
54 (NB 18 / MWB 17 / SWB 19)
267
1
181 (NB 59 / MWB 58 / SWB 64)
32 (NB 10 / MWB 10 / SWB 12)
54 (NB 18 / MWB 17 / SWB 19)
267
2
181 (NB 59 / MWB 58 / SWB 64)
33 (NB 11 / MWB 10 / SWB 12)
53 (NB 17 / MWB 17 / SWB 19)
267
3
181 (NB 59 / MWB 58 / SWB 64)
33 (NB 11 / MWB 10 / SWB 12)
53 (NB 17 / MWB 17 / SWB 19)
267
4
181 (NB 59 / MWB 58 / SWB 64)
33 (NB 11 / MWB 10 / SWB 12)
53 (NB 17 / MWB 17 / SWB 19)
267
Table 1
Generator
Discriminative score
Mean Wasserstein distance
Memorization ratio
SMOTE (tabular)
0.47 ± 0.06
0.26 ± 0.04
0.34 ± 0.06
BorderlineSMOTE (tabular)
0.57 ± 0.10
0.34 ± 0.07
0.37 ± 0.06
ADASYN (tabular)
0.42 ± 0.08
0.26 ± 0.05
0.35 ± 0.06
Gaussian (tabular)
1.00 ± 0.01
138.97 ± 25.32
1029.14 ± 242.48
WGAN-GP (tabular)
0.99 ± 0.02
0.80 ± 0.05
0.79 ± 0.07
Jitter (point cloud)
0.90 ± 0.14
1.44 ± 0.13
2.32 ± 0.37
Table 2
Generator
Data type
Times selected
Selection rate
BorderlineSMOTE
Tabular
4 / 10
40%
SMOTE
Tabular
3 / 10
30%
ADASYN
Tabular
2 / 10
20%
WGAN-GP
Tabular
1 / 10
10%
Gaussian (parametric)
Tabular
0 / 10
0%
Jitter augmentation
Point cloud
10 / 10
100%
Table 3
Model
Val Acc.
Val Prec.
Val Rec.
Val F1
Ensemble (XGBoost + MLP)
0.84 ± 0.04
0.84 ± 0.05
0.83 ± 0.04
0.83 ± 0.04
MLP
0.83 ± 0.08
0.85 ± 0.09
0.84 ± 0.09
0.83 ± 0.09
XGBoost
0.83 ± 0.06
0.84 ± 0.06
0.84 ± 0.07
0.83 ± 0.06
LightGBM
0.80 ± 0.10
0.81 ± 0.10
0.81 ± 0.10
0.81 ± 0.10
Table 4
Model
Test Acc.
Test 95% CI
Test Prec.
Test Rec.
Test F1
LightGBM
0.82 ± 0.05
[0.76, 0.88]
0.83 ± 0.04
0.82 ± 0.04
0.81 ± 0.05
Ensemble (XGBoost + MLP)
0.81 ± 0.05
[0.75, 0.88]
0.81 ± 0.06
0.81 ± 0.05
0.81 ± 0.06
XGBoost
0.80 ± 0.08
[0.71, 0.90]
0.81 ± 0.08
0.80 ± 0.07
0.79 ± 0.08
MLP
0.78 ± 0.09
[0.67, 0.90]
0.78 ± 0.10
0.78 ± 0.09
0.77 ± 0.10
Table 5
Model
Mean diff. vs. LightGBM (pp)
Paired t-test p-value
XGBoost
-1.90
0.485
MLP
+3.86
0.249
Ensemble
+0.86
0.777
Table 6
Model
Val Acc.
Val Prec.
Val Rec.
Val F1
Ensemble (Fusion-DGCNN+XGBoost+PGLR)
0.87 ± 0.01
0.86 ± 0.01
0.86 ± 0.01
0.86 ± 0.01
Fusion-PointBERTStyle
0.77 ± 0.08
0.78 ± 0.06
0.77 ± 0.08
0.77 ± 0.08
MLP
0.77 ± 0.11
0.80 ± 0.11
0.77 ± 0.11
0.77 ± 0.11
Fusion-PointNet
0.75 ± 0.10
0.77 ± 0.09
0.76 ± 0.10
0.75 ± 0.10
XGBoost
0.75 ± 0.11
0.78 ± 0.12
0.77 ± 0.11
0.76 ± 0.11
Fusion-PointNet++
0.74 ± 0.09
0.75 ± 0.08
0.74 ± 0.08
0.74 ± 0.09
Table 7
Model
Test Acc.
Test 95% CI
Test Prec.
Test Rec.
Test F1
Ensemble (FusionDGCNN+XGBoost+PGLR)
0.90 ± 0.03
[0.85, 0.93]
0.89 ± 0.04
0.88 ± 0.03
0.88 ± 0.03
LightGBM
0.84 ± 0.07
[0.75, 0.93]
0.86 ± 0.06
0.84 ± 0.07
0.84 ± 0.07
XGBoost
0.83 ± 0.04
[0.78, 0.88]
0.84 ± 0.04
0.82 ± 0.05
0.83 ± 0.04
Fusion-DGCNN
0.80 ± 0.08
[0.71, 0.90]
0.80 ± 0.10
0.79 ± 0.08
0.79 ± 0.09
PGLR
0.80 ± 0.09
[0.69, 0.92]
0.83 ± 0.08
0.80 ± 0.10
0.80 ± 0.09
Fusion-PointNet++
0.78 ± 0.11
[0.64, 0.93]
0.79 ± 0.12
0.78 ± 0.11
0.78 ± 0.12
Table 8
Model
Mean diff. vs. Ensemble (pp)
Paired t-test p-value
Fusion-PointNet
+13.57
0.015
Pure_PointNet
+57.67
<0.001
PointNet++
+11.52
0.063
Fusion-DGCNN
+9.62
0.049
PointBERT
+15.43
0.004
XGBoost
+6.62
0.008
Table 9
Fold
Full combo Test acc.
w/o Fusion-DGCNN
w/o XGBoost
w/o PGLR
1
0.86
−1.26
+3.79
+3.79
2
0.88
+6.56
+5.26
+7.88
3
0.90
+3.91
+5.22
+3.91
4
0.92
+6.48
+3.89
+3.89
5
0.94
−1.36
+1.36
+2.73
Table 10
Model
Mean marginal contribution (pp)
Std (pp)
PGLR
+4.44
1.99
XGBoost
+3.90
1.58
Fusion-DGCNN
+2.87
3.96
Table 11
Model
Mean contribution (pp)
2D+3D / 2D-only
LightGBM
+2.78 ± 0.64
2D+3D
PGLR
+2.36 ± 0.64
2D+3D
XGBoost
+2.08 ± 0.63
2D+3D
MLP
+1.31 ± 0.90
2D+3D
Fusion-PointBERTStyle
+0.07 ± 0.74
2D+3D
Fusion-DGCNN
+0.00 ± 0.57
2D+3D
Table 12
Rank
Combination
Size
Mean OOF accuracy
Marginal contribution within this combination (pp)
Department of Biological and Agricultural Engineering, University of Arkansas, Fayetteville, AR · Department of Food Science, University of Arkansas, Fayetteville, AR
Jun 15, 2026·Chirantan Sen Mukherjee, Seung-Chul Yoon, William J. Beksi
The authors are with the Department of Computer Science and Engineering, The University of Texas at Arlington, Arlington, TX, USA. · The author is with the Quality and Safety Assessment Research Unit, U.S. National Poultry Research Center, USDA Agricultural Research Service, Athens, GA, USA.
Department of Biological and Agricultural Engineering, University of Arkansas, Fayetteville, AR, USA, 72701. · Department of Food Science, University of Arkansas, Fayetteville, AR, USA, 72701. · Department of Poultry Science, University of Arkansas, Fayetteville, AR, USA, 72701.