(Occlusal-SEG) Articulating Paper Occlusal Contact Test Semantic Segmentation Dataset

THIS DATA IS ONLY TO BE USED FOR RESEARCH PURPOSES, IT IS NOT TO BE USED FOR COMMERCIAL PURPOSES Code can be found here: Github Occlusal-SEG is the accompanying dataset for our paper "H-FCBFormer: Hierarchical Fully Convolutional Branch Transformer for Occlusal Contact Segmentation with Articulating Paper". Please cite our original paper [1] (found below) and this repository, if this data is used. If the number of citations are a limitation, only cite our original paper. This data originated from Barcelona, Spain. Occlusal contact indication is a method used by dentists to determine the alignment of maxillary and mandibular teeth. To determine this alignment occlusal contact tests are carried out, to indicate the areas on upper and lower anterior teeth that make contact with each other. The most common test is Articulating Paper (AP), which marks the teeth with ink when pressure is applied (patient biting down), but it is often inaccurate. The best performing test is Occlufast Rock (OFR), which creates a silicone mould of the patients teeth indicating the occlusal contacts as areas of low silicone material, but it is too expensive for widespread clinical use. We propose this dataset to accompany our paper [1], that uses the original OFR and AP masks for the same patient, to indicate the areas of true and false contact indications, from a cropped digital image of a patient's teeth with AP ink indications. This dataset contains cropped .PNG images of patient teeth with AP test ink on teeth. Each image is accompanied with corresponding .PNG semantic mask annotations, containing Medically True Positive (MTP) annotations (pixel val: 255), Medically False Positive (MFP) annotations (pixel val: 254), and Background annotations (pixel val: 0). MTP masks are correct occlusal contact indications within the ink indicated area, MFP are incorrect occlusal contact indications, where the union of MTP and MFP is the full indicated AP area (original AP mask). MTP and MFP masks are created using gold standard OFR masks of the same patient and time of test. Details on the creation of these masks are found in our paper [1]. The dataset also contains gold standard OFR masks (no images) and original AP masks (no images) for evaluation and reference purposes. File Layout and Data Counts: File Name Description Data Counts Baseline_Data_Processed AP occlusal test images with accompanying .PNG semantic segmentation masks. This is our data using our proposed annotation methodology for MTP (pixel val: 255), MFP (pixel val: 254), and background (pixel val: 0) indications. This data is the full dataset without data splits. Images: 512 .png files [32 patients, 4 paper thicknesses, 2 Active/Passive application, 2 initial/secondary (Td/Rd) tests] Masks: 512 .png files [32 patients, 4 paper thicknesses, 2 Active/Passive application, 2 initial/secondary (Td/Rd) tests] Experiments_Processed AP occlusal test images with accompanying .PNG semantic segmentation masks. This is our data using our proposed annotation methodology for MTP (pixel val: 255), MFP (pixel val: 254), and background (pixel val: 0) indications. This data is the 4 fold split we used to train and evaluate our models. Each fold contains an alternating train/val split. Validation: 384 .png files (masks and images) using alternating train/val splits per 4 folds [24 patients, 4 paper thicknesses, 2 Active/Passive application, 2 initial/secondary (Td/Rd) tests] Test: 128 .png files (masks and images) using alternating train/val splits per 4 folds [8 patients, 4 paper thicknesses, 2 Active/Passive application, 2 initial/secondary (Td/Rd) tests] Gold_Standard_OFR_Masks Original OFR .PNG masks used to create our dataset. No digital images of the method are included. The masks include positive indications by the OFR test method (pixel val: 255), and negative indications by the OFR test (pixel val: 0). 256 .png files (masks only) for OFR test method [32 patients, 4 OFR sensitivities, 2 initial/secondary (Td/Rd) tests] Original_AP_Masks Original AP .PNG masks used to create our dataset. No digital images of the method are included in this folder, but can be found in the "Baseline_Data_Processed" and "Experiments_Processed" folder. The masks include positive indications by the AP test method (pixel val: 255), and negative indications by the AP test (pixel val: 0). 512 .png files (masks only) for AP test method [32 patients, 4 paper thicknesses, 2 Active/Passive application, 2 initial/secondary (Td/Rd) tests] The original collection contained 35 patients. However, 3 patients were removed due to annotation/test errors. These patients are patient 20, 26 and 35. Data Naming Convention: Our naming convention is: Z01Td-AP012A.png where: Z[Patient][Test/Retest]-[Test Method][Thickness][Active/Passive].png for AP: Patient: [01/02/03/04/05/06/07/08/09/10/11/12/13/14/15/16/17/18/19/21/22/23/24/25/27/28/29/30/31/32/33/34] Test/Retest: [Td/Rd] Test Method: [AP] Thickness: [012/040/100/200] Active/Passive: [A/P] for OFR: Z01Td-OFR40.png Patient: [01/02/03/04/05/06/07/08/09/10/11/12/13/14/15/16/17/18/19/21/22/23/24/25/27/28/29/30/31/32/33/34] Test/Retest: [Td/Rd] Test Method: [OFR] Thickness: [40/50/100/200] Active/Passive: NONE Bibtex For Original Paper: [1] @InProceedings{banks2024hierarchical, author="Banks, Ryan and Rovira-Lastra, Bernat and Martinez-Gomis, Jordi and Chaurasia, Akhilanand and Li, Yunpeng", editor="Yap, Moi Hoon and Kendrick, Connah and Behera, Ardhendu and Cootes, Timothy and Zwiggelaar, Reyer", title="H-FCBFormer: Hierarchical Fully Convolutional Branch Transformer for Occlusal Contact Segmentation with Articulating Paper", booktitle="Medical Image Understanding and Analysis", year="2024", publisher="Springer Nature Switzerland", pages="72--86", doi="10.1007/978-3-031-66958-3_6", isbn="978-3-031-66958-3" }