task_categories:
- image-classification
language:
- en
pretty_name: DMR-IR Thermal Breast Cancer Dataset
size_categories:
- 1K<n<10K
tags:
- medical
configs:
- config_name: default
data_files:
- split: train
path: with_mastectomy/train-*
- split: validation
path: with_mastectomy/validation-*
- split: test
path: with_mastectomy/test-*
dataset_info:
- config_name: default
features:
- name: image
dtype: image
- name: label
dtype:
class_label:
names:
'0': benign
'1': malignant
- name: text
dtype: string
- name: patient_id
dtype: string
- name: text_embedding
list: float32
- name: segmentation_mask
list:
list:
list: uint8
- name: protocol
dtype: string
- name: view
dtype:
class_label:
names:
'0': Frontal
'1': Right 45°
'2': Right 90°
'3': Left 45°
'4': Left 90°
'5': Unknown
- name: record
dtype: string
- name: role
dtype: string
- name: age_current
dtype: int64
- name: age_at_visit
dtype: int64
- name: registration_date
dtype: string
- name: marital_status
dtype:
class_label:
names:
'0': Divorced
'1': Married
'2': Single
'3': Widow
- name: race
dtype:
class_label:
names:
'0': Asian
'1': Black
'2': Indigenous
'3': Mulatto
'4': Multiracial
'5': White
- name: visit_date
dtype: string
- name: complaints
dtype: string
- name: symptoms
dtype: string
- name: signs
dtype: string
- name: last_menstrual_period
dtype: string
- name: menopause
dtype: string
- name: menarche
dtype: int64
- name: eating_habits
dtype:
class_label:
names:
'0': '35.50'
'1': High in fat
'2': Low in fat
'3': No fat
- name: cancer_family
dtype: string
- name: family_history
dtype: string
- name: further_informations
dtype: string
- name: mammography
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: radiotherapy
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: plastic_surgery
dtype:
class_label:
names:
'0': 'No'
'1': Yes, both breasts
'2': Yes, left breast
'3': Yes, right breast
- name: prosthesis
dtype:
class_label:
names:
'0': 'No'
'1': Yes, both breasts
'2': Yes, left breast
'3': Yes, right breast
- name: biopsy
dtype:
class_label:
names:
'0': 'No'
'1': Yes, both breasts
'2': Yes, left breast
'3': Yes, right breast
- name: use_of_hormone_replacement
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: nipple_changes
dtype:
class_label:
names:
'0': Yes, both breasts
'1': Yes, left breast
'2': Yes, right breast
- name: is_there_signal_of_wart_on_breast
dtype:
class_label:
names:
'0': 'No'
'1': Yes, both breasts
'2': Yes, left breast
'3': Yes, right breast
- name: medical_further_informations
dtype: string
- name: body_temperature
dtype: float64
- name: protocol_smoked
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: protocol_drank_coffee
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: protocol_consumed_alcohol
dtype:
class_label:
names:
'0': 'No'
- name: protocol_physical_exercise
dtype:
class_label:
names:
'0': 'No'
- name: >-
protocol_put_some_pomade_deodorant_or_products_at_breasts_or_armpits_region
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
- name: mastectomy
dtype:
class_label:
names:
'0': 'No'
'1': 'Yes'
splits:
- name: train
num_bytes: 6262191255.51803
num_examples: 4874
- name: validation
num_bytes: 1259102362.9086394
num_examples: 980
- name: test
num_bytes: 1315635084.928
num_examples: 1024
download_size: 4969929658
dataset_size: 8836928703.35467
Dataset Card for DMR-IR
DMR-IR is an infrared imaging dataset for Mamma Research, featuring TIFF raw temperature images and clinical metadata with generated text prompts from data acquired at Antônio Pedro University Hospital. It supports multimodal research in breast imaging and diagnostics.
Dataset Details
Dataset Description
The DMR-IR dataset is an infrared imaging database for Mamma Research, developed from clinical data acquired at the Antônio Pedro University Hospital. It comprises IR images, digitized mammograms, and associated clinical data from patients in both the screening and gynecologic departments. The IR images were captured using a FLIR SC620 thermal camera with a sensitivity of less than 0.04°C and a capture range of -40°C to 500°C, at a resolution of 640 × 480 pixels. Data acquisition was approved by the Ethical Committee and registered with the Brazilian Ministry of Health (CAAE: 01042812.0.0000.5243). The dataset is accessible through an online interface.
In our processed version, the original temperature matrices have been converted to TIFF format. Additionally, clinical metadata has been used to generate descriptive text prompts via a custom Python function, summarizing personal, clinical, and protocol information to facilitate multimodal research.
- Curated by: Guillermo Pinto, Brayan Quintero, Julian Leon, Miguel Pimiento, Dana Villamizar
- Shared by [optional]: The original dataset was shared by Silva, L. F.; Saade, D. C. M.; Sequeiros, G. O.; Silva, A. C.; Paiva, A. C.; Bravo, R. S.; Conci, A.
- Language(s) (NLP): English.
- License: Not provided – please update if applicable.
Dataset Sources [optional]
- Repository: DMR-IR Online Interface
- Paper [optional]: DOI: https://doi.org/10.1166/jmihi.2014.1226
Uses
Direct Use
This dataset is intended for research in breast imaging and diagnostic analysis, including:
- Developing and testing IR image processing and analysis algorithms.
- Investigating clinical correlations with infrared imaging features.
- Utilizing generated text prompts for multimodal research.
Out-of-Scope Use
- The dataset should not be used as the sole source for clinical decision making without extensive validation.
- It is not recommended for commercial applications if license restrictions apply.
- Applications outside of its intended research context are not advised.
Dataset Creation
Curation Rationale
The dataset was curated to improve the usability of the original IR images by converting them from temperature matrices to TIFF format, enhancing reproducibility. In addition, the integration of descriptive clinical metadata allows the clinical context to be incorporated into image analysis workflows. Furthermore, the dataset is not locked, meaning its contents may evolve over time. For research purposes—particularly when the dataset is used to conduct a study—it is necessary to work with a locked version to ensure reproducibility and consistency of the findings. Additionally, some data anomalies were corrected following the recommendations presented in https://doi.org/10.1177/14604582231153779.
Source Data
The original data was collected at Antônio Pedro University Hospital and comprises:
- Infrared images captured with a FLIR SC620 thermal camera.
- Digitized mammograms.
- Clinical data from both healthy patients and patients with breast diseases.
Data Collection and Processing
- Original Collection: Data was acquired from patients in the screening and gynecologic departments following ethical guidelines approved by the hospital’s Ethical Committee and registered under CAAE: 01042812.0.0000.5243.
Annotations [optional]
Annotation process
The dataset does not include manual annotations beyond the automated generation of clinical text prompts based on existing metadata. Only the mastectomy column was manually annotated, while the remaining metadata fields were cleaned and corrected. The full pipeline to reproduce the dataset creation is available at https://github.com/semilleroCV/BreastCATT/tree/main/data/DMR-IR.
Personal and Sensitive Information
The clinical metadata includes sensitive personal and medical information such as risk factors (e.g., eating habits, family history, radiotherapy, hormone replacement, age, menarche/menopause), complementary features (e.g., prosthesis, nipple changes), and protocol details (e.g., smoking, alcohol, coffee consumption). Although explicit identifiers are not provided, the data is sensitive and should be handled in accordance with applicable privacy regulations.
Bias, Risks, and Limitations
- The dataset reflects the clinical demographics and practices of a specific hospital in Brazil, which may limit its generalizability to other populations.
- The calibration process and text prompt generation depend on the accuracy of the thermal parser tool and the quality of the clinical metadata.
- The selection of clinical features may introduce bias that could affect downstream analysis.
Additional Information
Citation Information
Dataset paper
@article{silva2014new,
title={A new database for breast research with infrared image},
author={Silva, LF and Saade, DCM and Sequeiros, GO and Silva, AC and Paiva, AC and Bravo, R de S and Conci, Aura},
journal={Journal of Medical Imaging and Health Informatics},
volume={4},
number={1},
pages={92--100},
year={2014},
publisher={American Scientific Publishers}
}
Our research paper
@INPROCEEDINGS{pintobreastcatt,
author={Pinto, Guillermo and León, Julián and Quintero, Brayan and Villamizar, Dana and Rueda-Chacón, Hoover},
booktitle={2025 IEEE Colombian Conference on Applications of Computational Intelligence (ColCACI)},
title={Multimodal Vision-Language Transformer for Thermography Breast Cancer Classification},
year={2025},
volume={},
number={},
pages={1-6},
keywords={Sensitivity;Translation;Mortality;Infrared imaging;Medical services;Metadata;Transformers;Breast cancer;Standards;Periodic structures;Breast cancer;deep learning;thermography;vision-language transformer;clinical metadata;cross-attention},
doi={10.1109/ColCACI67437.2025.11230909}}