Artificial Intelligence for Locally Advanced Head Neck Cancer Treated With Multi-modality Adaptive RadioTherapy: Machine Learning-based Radiomic Prediction of Outcome and Toxicity (RadiomicART)
Artificial Intelligence for Locally Advanced Head Neck Cancer Treated With Multi-modality Adaptive RadioTherapy: Machine Learning-based Radiomic Prediction of Outcome and Toxicity (RadiomicART)
Current clinical management algorithms for squamous cell carcinoma of head and neck (HNSCC) involve the use of surgery and / or radiotherapy (RT) depending on the stage of the disease at diagnosis. Radical RT, exclusive or in combination with systemic therapy, represents an effective therapeutic option according to the international guidelines. Despite the recent technological advancements in the field of RT, about 30-50% of patients will develop locoregional failure after primary treatment . Moreover, although the development of Intensity modulated radiation therapy (IMRT) and Volumetric modulated arc therapy (VMAT) techniques allowed a greater sparing of dose on healthy tissues, radiation-induced toxicity still represents a relevant concern, impacting on quality of life. The continuous effort of personalized medicine has the goal of improving patient's outcome, in terms of both disease's control and pattern of toxicity. Advanced imaging modalities appear to play an essential role in the customization of the radiation treatment as shown through the use of Adaptive Radiotherapy (ART) and radiomic. With ART we mean the adaptation of tumor volumes and surrounding organs at risk (OARs) to the shrinkage and patient emaciation during RT treatment. Adaptive radiotherapy (ART) includes techniques that allow knowledge of patient-specific anatomical variations informed by Image-guided radiotherapies (IGRTs) to feedback into the plan and dose-delivery optimization during the treatment course. Radiomic is the extraction of quantitative features from medical images to characterize tumor pathology or heterogeneity. Radiomic features extracted from medical images can be used as input features to create a machine learning model able to predict survival, and to guide treatment thanks to its predictive value in view of therapy personalization.
The combination of both ART and radiomic analysis could potentially be considered a further advance in the personalization of oncological treatments, and in particular for radiation treatments. For this reason, the investigators designed the present research project with the aim to prospectively evaluate a machine learning-based radiomic approach to predict outcome and toxicity of HNSCC patients treated with ART by mean of CT, MRI and PET-scan.
Squamous cell carcinoma of the head and neck (HNSCC) is characterized by an incidence in Europe of 140.000 new cases per year, with survival rates at 5 years ranging from 25 to 65%.
Current clinical management algorithms for HNSCC patients involve the use of surgery and / or radiotherapy depending on the stage of the disease at diagnosis. Radical radiotherapy (RT), exclusive or in combination with systemic therapy, represents an effective therapeutic option according to the international guidelines.
Despite the recent technological advancements in the field of radiation therapy, about 30-50% of patients will develop locoregional failure after primary treatment of head and neck cancer. Moreover, although the development of Intensity modulated radiation therapy (IMRT) and Volumetric modulated arc therapy (VMAT) techniques allowed a greater sparing of dose on healthy tissues, radiation-induced toxicity still represents a relevant concern, impacting on quality of life of cancer patients even for long time after treatment.
The continuous effort of personalized medicine has the goal of improving patient's outcome, in terms of both disease's control and pattern of toxicity. Advanced imaging modalities appear to play an essential role in the customization of the radiation treatment as shown through the use of Adaptive Radiotherapy (ART) and radiomic.
With ART the investigators mean the adaptation of tumor volumes and surrounding organs at risk (OARs) to the shrinkage and patient emaciation during RT treatment.
The recent literature showed that tumor shrinkage can reach 70% by the end of the RT treatment, and at the same time OARs, such as parotid glands, can reduce their size by 7 to 70%. These alterations, if not taken into account, can lead to an unexpected delivery of lower dose on the tumor and higher dose of OARs compared to what planned.
Adaptive radiotherapy (ART) includes techniques that allow knowledge of patient-specific anatomical variations informed by Image-guided radiotherapies (IGRTs) to feedback into the plan and dose-delivery optimization during the treatment course. The persisting weakest link in the treatment chain for radiotherapy remains clinician-led target identification.
Compared to CT or CBCT, MRI offers superior soft-tissue definition with no associated radiation risk. MRI identifies targets larger than on CT because tumour that otherwise would have been missed is now seenah; however, most commonly, targets are reported to be smaller when delineated on MRI. The resulting smaller MRI-derived target improves the therapeutic ratio so enabling dose escalation. The availability of 'functional' MRI sequences holds promise that geometric adaptation maybe complemented by biological adaptation. Diffusion-weighted imaging (DWI) is a functional imaging technique dependent on the random motion of water molecules to generate image contrast. As tumours usually have greater cellularity than normal tissue, they demonstrate higher signal intensity, i.e., restricted diffusion on MRI. This is reflected in the low mean apparent diffusion coefficient (ADC) value. This has potential to provide both qualitative and quantitative information. Change in the ADC has been used to identify early treatment response, and to predict local recurrence. Therefore, on-board DWI could identify early non-responders who may benefit from change in treatment approach.
Radiomic is the extraction of quantitative features from medical images to characterize tumor pathology or heterogeneity. Radiomic features extracted from medical images can be used as in put features to create a machine learning model able to predict survival, and to guide treatment thanks to its predictive value in view of therapy personalization.
We previously evaluated in a retrospective study the qualitative analysis of the radiomic characteristics of head and neck tumor tissues, in order to identify a predictive signature of the biological characteristics of the tumor. The investigators stratified HNSCC patients according to the most significant radiomic features into high- and low-risk groups of relapse and survival after radio-chemotherapy.
The combination of both ART and radiomic analysis could potentially be considered a further advance in the personalization of oncological treatments, and in particular for radiation treatments. For this reason, the investigator designed the present research project with the aim to prospectively evaluate a machine learning-based radiomic approach to predict outcome and toxicity of HNSCC patients treated with ART by mean of CT, MRI and PET-scan.
Inclusion Criteria:
Exclusion Criteria: