File size: 6,648 Bytes
f176c40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6051c52
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
---
license: apache-2.0
datasets:
  - jeffrey423/ToothXpert.MM-OPG-Annotations
language:
  - en
tags:
  - dental
  - medical
  - multimodal
  - vision-language
  - llava
  - clip
  - sam
  - lora
  - orthopantomography
  - opg
  - x-ray
  - diagnosis
base_model: liuhaotian/llava-v1.5-7b
pipeline_tag: image-text-to-text
library_name: transformers
---

# ToothXpert Model

ToothXpert is a multimodal AI model for comprehensive dental X-ray (OPG) analysis, combining vision and language understanding for automatic diagnosis and condition detection.

## Quick Start

### Installation

```bash
pip install torch torchvision transformers
pip install opencv-python einops peft medpy
pip install "numpy<2.0"  # Important for compatibility
```

### Download Model

```python
from huggingface_hub import snapshot_download

model_path = snapshot_download(
    repo_id='jeffrey423/ToothXpert',
    local_dir='./ToothXpert_pretrained'
)
```

### Simple Inference

```python
import cv2
import torch
import torch.nn.functional as F
from transformers import AutoTokenizer, CLIPImageProcessor
from model.ToothXpert_MOE import ToothXpertForCausalLMMOE
from model.llava import conversation as conversation_lib
from model.llava.mm_utils import tokenizer_image_token
from model.segment_anything.utils.transforms import ResizeLongestSide
from utils.utils import (DEFAULT_IM_END_TOKEN, DEFAULT_IM_START_TOKEN,
                         DEFAULT_IMAGE_TOKEN, IMAGE_TOKEN_INDEX)

# Preprocessing function
def preprocess(x, pixel_mean=torch.Tensor([123.675, 116.28, 103.53]).view(-1, 1, 1),
               pixel_std=torch.Tensor([58.395, 57.12, 57.375]).view(-1, 1, 1), img_size=1024):
    x = (x - pixel_mean) / pixel_std
    h, w = x.shape[-2:]
    padh = img_size - h
    padw = img_size - w
    x = F.pad(x, (0, padw, 0, padh))
    return x

# Load model
model_path = "./ToothXpert_pretrained"
device = "cuda:0"

tokenizer = AutoTokenizer.from_pretrained(
    model_path,
    model_max_length=512,
    padding_side="right",
    use_fast=False,
)
tokenizer.pad_token = tokenizer.unk_token
tokenizer.add_tokens("[SEG]")
seg_token_idx = tokenizer("[SEG]", add_special_tokens=False).input_ids[0]
tokenizer.add_tokens([DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN], special_tokens=True)

moe_lora_args = {
    "lora_r": 8,
    "lora_alpha": 16,
    "lora_dropout": 0.05,
    "lora_target_modules": "q_proj,v_proj",
    "moe_lora": False,
    "expert_num": 3,
    "guide": True,
    "guide_mode": "smmulsm",
    "vocab_size": len(tokenizer),
}

model = ToothXpertForCausalLMMOE.from_pretrained(
    model_path,
    low_cpu_mem_usage=True,
    vision_tower="openai/clip-vit-large-patch14",
    seg_token_idx=seg_token_idx,
    torch_dtype=torch.bfloat16,
    train_mask_decoder=True,
    out_dim=256,
    moe_lora_args=moe_lora_args,
)

model.config.eos_token_id = tokenizer.eos_token_id
model.config.bos_token_id = tokenizer.bos_token_id
model.config.pad_token_id = tokenizer.pad_token_id

model.get_model().initialize_vision_modules(model.get_model().config)
vision_tower = model.get_model().get_vision_tower()
vision_tower.to(dtype=torch.bfloat16, device=device)

model = model.bfloat16().to(device)
model.eval()

# Load and process image
image_path = "your_dental_xray.png"
image_np = cv2.imread(image_path)
image_np = cv2.cvtColor(image_np, cv2.COLOR_BGR2RGB)
original_size_list = [image_np.shape[:2]]

clip_image_processor = CLIPImageProcessor.from_pretrained("openai/clip-vit-large-patch14")
transform = ResizeLongestSide(1024)

image_clip = (
    clip_image_processor.preprocess(image_np, return_tensors="pt")["pixel_values"][0]
    .unsqueeze(0).to(device).bfloat16()
)

image = transform.apply_image(image_np)
resize_list = [image.shape[:2]]
image = (
    preprocess(torch.from_numpy(image).permute(2, 0, 1).contiguous())
    .unsqueeze(0).to(device).bfloat16()
)

# Prepare prompt
question = "Can you describe the image for me?"
conv = conversation_lib.conv_templates["llava_v1"].copy()
conv.messages = []
prompt = DEFAULT_IMAGE_TOKEN + "\n" + question
prompt = prompt.replace(DEFAULT_IMAGE_TOKEN,
                       DEFAULT_IM_START_TOKEN + DEFAULT_IMAGE_TOKEN + DEFAULT_IM_END_TOKEN)

conv.append_message(conv.roles[0], prompt)
conv.append_message(conv.roles[1], "")
prompt = conv.get_prompt()

input_ids = tokenizer_image_token(prompt, tokenizer, return_tensors="pt")
input_ids = input_ids.unsqueeze(0).to(device)

# Run inference
with torch.no_grad():
    output_ids, pred_masks = model.evaluate(
        image_clip,
        image,
        input_ids,
        resize_list,
        original_size_list,
        max_new_tokens=512,
        tokenizer=tokenizer,
    )

output_ids = output_ids[0][output_ids[0] != IMAGE_TOKEN_INDEX]
text_output = tokenizer.decode(output_ids, skip_special_tokens=False)
text_output = text_output.split('ASSISTANT:')[-1].replace('</s>', '').strip()

print(f"Question: {question}")
print(f"Answer: {text_output}")
```

## Example Questions

**General Description:**
- "Can you describe the image for me?"

**Specific Conditions:**
- "Is there any amalgam restorations in the image?"
- "Any R/L suggestive of caries present?"
- "Is there any dental implant present?"
- "Is there any root canal treated teeth?"

## Supported Conditions

ToothXpert can detect 11 dental conditions:
1. Amalgam restorations
2. Caries (R/L)
3. Crestal bone loss (mandible)
4. Crestal bone loss (maxillary)
5. Implant-supported bridge
6. Dental implant
7. Metallic/non-metallic post
8. Non-metallic restorations
9. Periapical radiolucency
10. Root canal treated teeth
11. Tooth-supported bridge

## Requirements

- **GPU**: NVIDIA GPU with at least 16GB VRAM
- **Python**: 3.11 (recommended)
- **CUDA**: 12.1 or compatible

## Model Details

- **Base Model**: LLaVA-1.5-7B
- **Vision Encoder**: CLIP ViT-L/14
- **Segmentation**: SAM (Segment Anything Model) ViT-H
- **Adaptation**: Guided Mixture of LoRA Experts (G-MoLE)
- **Model Size**: ~15GB

## Citation

If you use ToothXpert in your research, please cite:

```bibtex
@article{liu2026toothxpert,
  title={Developing and Evaluating Multimodal Large Language Model for Orthopantomography Analysis to Support Clinical Dentistry},
  author={Liu, Xinyu and Hung, Kuo Feng and Yu, Weihao and Ng, Ray Anthony W T and Li, Wuyang and Niu, Tianye and Chen, Hui and Yuan, Yixuan},
  journal={Cell Reports Medicine},
  year={2026}
}
```

## Links

- **GitHub Repository**: [CUHK-AIM-Group/ToothXpert](https://github.com/CUHK-AIM-Group/ToothXpert)
- **Dataset**: [jeffrey423/ToothXpert.MM-OPG-Annotations](https://huggingface.co/datasets/jeffrey423/ToothXpert.MM-OPG-Annotations)

## License

Apache License 2.0