| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 3.87 MB | Adobe PDF |
Advisor(s)
Abstract(s)
O cancro oral, particularmente o carcinoma oral das células escamosas (COCE), é um dos tipos
mais prevalentes e letais de cancro, representando cerca de 95% dos cancros orais. O
diagnóstico precoce do COCE é essencial para aumentar a taxa de sobrevivência dos pacientes,
sendo tradicionalmente realizado através da análise de imagens histopatológicas por
profissionais de saúde. No entanto, o uso de técnicas de Inteligência Artificial, como o Deep
Learning, tem o potencial de revolucionar este processo, permitindo diagnósticos mais rápidos
e precisos.
Nesta investigação, foi desenvolvido um modelo de Deep Learning baseado na arquitetura
EfficientNetB3, combinado com camadas de Transformer para melhorar a extração de
características e a capacidade de atenção do modelo. Este modelo foi treinado utilizando
técnicas de Transfer Learning, que permitiram aproveitar os conhecimentos adquiridos com
grandes conjuntos de dados previamente treinados, adaptando-os ao problema específico de
classificação de imagens histopatológicas. Foram também implementadas técnicas de aumento
de dados e normalização para aumentar a robustez do modelo.
Os resultados obtidos mostraram que o modelo proposto alcançou uma taxa de acerto de 96%
no conjunto de teste, superando modelos existentes. Estes resultados evidenciam a eficácia do
modelo na classificação precisa de tecido saudável e cancerígeno em imagens histopatológicas.
Além disso, foi realizada uma análise estatística para verificar a significância dos resultados
obtidos, confirmando a superioridade do modelo desenvolvido.
No futuro, propõe-se a adaptação do modelo para a deteção de subtipos de cancro oral e a
utilização de técnicas de segmentação de imagens, que poderão permitir uma deteção mais
precisa das áreas afetadas pelo cancro, complementando o processo de diagnóstico clínico. A
integração deste tipo de modelos num sistema de apoio à decisão clínica é também uma direção
promissora, com o potencial de melhorar o tempo e a precisão do diagnóstico em ambiente
hospitalar.
Oral cancer, particularly oral squamous cell carcinoma (OSCC), is one of the most prevalent and lethal types of cancer, accounting for around 95% of oral cancers. Early diagnosis of COCE is essential to increase the survival rate of patients and has traditionally been carried out through the analysis of histopathological images by health professionals. However, the use of Artificial Intelligence techniques, such as Deep Learning, has the potential to revolutionize this process, allowing for faster and more accurate diagnoses. In this research, a Deep Learning model was developed based on the EfficientNetB3 architecture, combined with Transformer layers to improve feature extraction and the model's attention capacity. This model was trained using Transfer Learning techniques, which made it possible to take advantage of the knowledge acquired from large sets of previously trained data, adapting it to the specific problem of classifying histopathological images. Data augmentation and normalization techniques were also implemented to increase the model's robustness. The results obtained showed that the proposed model achieved a 96% hit rate on the test set, outperforming existing models. These results demonstrate the model's effectiveness in accurately classifying healthy and cancerous tissue in histopathology images. In addition, a statistical analysis was carried out to verify the significance of the results obtained, confirming the superiority of the model developed. In the future, it is proposed to adapt the model for the detection of oral cancer subtypes and the use of image segmentation techniques, which could allow more accurate detection of areas affected by cancer, complementing the clinical diagnosis process. Integrating this type of model into a clinical decision support system is also a promising direction, with the potential to improve the time and accuracy of diagnosis in a hospital environment.
Oral cancer, particularly oral squamous cell carcinoma (OSCC), is one of the most prevalent and lethal types of cancer, accounting for around 95% of oral cancers. Early diagnosis of COCE is essential to increase the survival rate of patients and has traditionally been carried out through the analysis of histopathological images by health professionals. However, the use of Artificial Intelligence techniques, such as Deep Learning, has the potential to revolutionize this process, allowing for faster and more accurate diagnoses. In this research, a Deep Learning model was developed based on the EfficientNetB3 architecture, combined with Transformer layers to improve feature extraction and the model's attention capacity. This model was trained using Transfer Learning techniques, which made it possible to take advantage of the knowledge acquired from large sets of previously trained data, adapting it to the specific problem of classifying histopathological images. Data augmentation and normalization techniques were also implemented to increase the model's robustness. The results obtained showed that the proposed model achieved a 96% hit rate on the test set, outperforming existing models. These results demonstrate the model's effectiveness in accurately classifying healthy and cancerous tissue in histopathology images. In addition, a statistical analysis was carried out to verify the significance of the results obtained, confirming the superiority of the model developed. In the future, it is proposed to adapt the model for the detection of oral cancer subtypes and the use of image segmentation techniques, which could allow more accurate detection of areas affected by cancer, complementing the clinical diagnosis process. Integrating this type of model into a clinical decision support system is also a promising direction, with the potential to improve the time and accuracy of diagnosis in a hospital environment.
Description
Keywords
Oral cancer Histopathology Deep learning CNN Image classification Transfer learning Transformer Cancro oral Histopatologia Classificação de imagens