Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification
Abstract Accurate diagnosis of oral lesions, early indicators of oral cancer, is a complex clinical challenge. Recent advances in deep learning have demonstrated potential in supporting clinical decisions. This paper introduces a deep learning model for classifying oral lesions, focusing on accuracy...
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2024-12-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-024-81724-0 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841559407788294144 |
|---|---|
| author | Adeetya Patel Camille Besombes Theerthika Dillibabu Mridul Sharma Faleh Tamimi Maxime Ducret Peter Chauvin Sreenath Madathil |
| author_facet | Adeetya Patel Camille Besombes Theerthika Dillibabu Mridul Sharma Faleh Tamimi Maxime Ducret Peter Chauvin Sreenath Madathil |
| author_sort | Adeetya Patel |
| collection | DOAJ |
| description | Abstract Accurate diagnosis of oral lesions, early indicators of oral cancer, is a complex clinical challenge. Recent advances in deep learning have demonstrated potential in supporting clinical decisions. This paper introduces a deep learning model for classifying oral lesions, focusing on accuracy, interpretability, and reducing dataset bias. The model integrates three components: (i) a Classification Stream, utilizing a CNN to categorize images into 16 lesion types (baseline model), (ii) a Guidance Stream, which aligns class activation maps with clinically relevant areas using ground truth segmentation masks (GAIN model), and (iii) an Anatomical Site Prediction Stream, improving interpretability by predicting lesion location (GAIN+ASP model). The development dataset comprised 2765 intra-oral digital images of 16 lesion types from 1079 patients seen at an oral pathology clinic between 1999 and 2021. The GAIN model demonstrated a 7.2% relative improvement in accuracy over the baseline for 16-class classification, with superior class-specific balanced accuracy and AUC scores. Additionally, the GAIN model enhanced lesion localization and improved the alignment between attention maps and ground truth. The proposed models also exhibited greater robustness against dataset bias, as shown in ablation studies. |
| format | Article |
| id | doaj-art-00ed48bd18d946d2abf72da7b4cccb46 |
| institution | Kabale University |
| issn | 2045-2322 |
| language | English |
| publishDate | 2024-12-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-00ed48bd18d946d2abf72da7b4cccb462025-01-05T12:28:38ZengNature PortfolioScientific Reports2045-23222024-12-0114111210.1038/s41598-024-81724-0Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classificationAdeetya Patel0Camille Besombes1Theerthika Dillibabu2Mridul Sharma3Faleh Tamimi4Maxime Ducret5Peter Chauvin6Sreenath Madathil7Faculty of Dental Medicine and Oral Health Sciences, McGill UniversityFaculty of Dental Medicine and Oral Health Sciences, McGill UniversityFaculty of Dental Medicine and Oral Health Sciences, McGill UniversityFaculty of Dental Medicine and Oral Health Sciences, McGill UniversityCollege of Dental Medicine, QU Health, Qatar UniversityFaculté d’Odontologie, Université Claude Bernard Lyon 1Faculty of Dental Medicine and Oral Health Sciences, McGill UniversityFaculty of Dental Medicine and Oral Health Sciences, McGill UniversityAbstract Accurate diagnosis of oral lesions, early indicators of oral cancer, is a complex clinical challenge. Recent advances in deep learning have demonstrated potential in supporting clinical decisions. This paper introduces a deep learning model for classifying oral lesions, focusing on accuracy, interpretability, and reducing dataset bias. The model integrates three components: (i) a Classification Stream, utilizing a CNN to categorize images into 16 lesion types (baseline model), (ii) a Guidance Stream, which aligns class activation maps with clinically relevant areas using ground truth segmentation masks (GAIN model), and (iii) an Anatomical Site Prediction Stream, improving interpretability by predicting lesion location (GAIN+ASP model). The development dataset comprised 2765 intra-oral digital images of 16 lesion types from 1079 patients seen at an oral pathology clinic between 1999 and 2021. The GAIN model demonstrated a 7.2% relative improvement in accuracy over the baseline for 16-class classification, with superior class-specific balanced accuracy and AUC scores. Additionally, the GAIN model enhanced lesion localization and improved the alignment between attention maps and ground truth. The proposed models also exhibited greater robustness against dataset bias, as shown in ablation studies.https://doi.org/10.1038/s41598-024-81724-0Oral lesion diagnosisInterpretabilityGuided attention inference networkBias mitigationCNN |
| spellingShingle | Adeetya Patel Camille Besombes Theerthika Dillibabu Mridul Sharma Faleh Tamimi Maxime Ducret Peter Chauvin Sreenath Madathil Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification Scientific Reports Oral lesion diagnosis Interpretability Guided attention inference network Bias mitigation CNN |
| title | Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification |
| title_full | Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification |
| title_fullStr | Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification |
| title_full_unstemmed | Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification |
| title_short | Attention-guided convolutional network for bias-mitigated and interpretable oral lesion classification |
| title_sort | attention guided convolutional network for bias mitigated and interpretable oral lesion classification |
| topic | Oral lesion diagnosis Interpretability Guided attention inference network Bias mitigation CNN |
| url | https://doi.org/10.1038/s41598-024-81724-0 |
| work_keys_str_mv | AT adeetyapatel attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT camillebesombes attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT theerthikadillibabu attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT mridulsharma attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT falehtamimi attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT maximeducret attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT peterchauvin attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification AT sreenathmadathil attentionguidedconvolutionalnetworkforbiasmitigatedandinterpretableorallesionclassification |