Abstract

This article provides an overview for primary cardiac sarcoma in regards to diagnostics, treatments and use of artificial intelligence (AI) as a possible tool. In this review, we conclude that primary cardiac sarcomas are exceptionally rare malignant tumours, localised in either the left or right chamber of the heart. Due to its rarity, limited cases and minimal surgical experiences for primary cardiac sarcoma are reported. Consequently, symptoms are frequently misinterpreted as heart failure or other cardiac diseases, delaying diagnostics and surgical management. While surgical resection remains the best treatment option, it is a major challenge for surgeons due to the tumour’s anatomic location and heterogeneity. However, investigation regarding the use of AI for higher diagnostic accuracy, treatment management and surgical assessments is becoming increasingly relevant. Ongoing developments suggest that AI tools can be utilised in the foreseeable future to aid with individualised treatment for the complexity of primary cardiac sarcomas, ultimately improving prognosis and patient outcomes.   

1. Introduction

Primary cardiac sarcoma (PCS) is a rare disease with aggressive malignancy that has a reported five-year survival rate of 11.5% (Guan et al., 2022). The low prognosis comes from the lack of experience that professionals have in the diagnosis and treatment of this cancer type (Ugarte et al., 2021; Hoffmeier et al., 2014). This condition blocks the blood flow and can lead to an irregular heart rate which can ultimately cause rapid heart failure, illustrating  how fatal this condition can quickly become and therefore demonstrating the high risks associated with it (Hoffmeier et al., 2014). One of the main reasons as to why this condition is incredibly difficult to treat is due to metastasis (the formation of secondary tumours) which obstructs efficient treatment of the disease. Approximately one quarter of all cardiac tumours are identified to be malignant (cancerous tumours that have the ability to metastasise and spread around the body) which is mainly due to mesenchymal stem cells. These stem cells have the ability to renew themselves and undergo multilineage differentiation which means they can specialise into many different types of cells (Sun et al., 2018; Ding et al., 2011).

The diagnosis of PCS is difficult due to how the symptoms can be frequently mistaken for other heart conditions, such as blood clots (Shanmugam, 2006; Tyebally et al., 2020). Some of these symptoms include high temperatures, weight loss, aching muscles and fatigue. The process undertaken to diagnose a patient contains multiple steps. Usually, a patient is first diagnosed after they suffer a stroke and an echocardiography is taken. However, if the tumour cannot be verified through this scan, a tomography is requested to confirm whether the tumour is benign or malignant. In some instances, angiography or CT scans can be used; specifically, angiographies are useful when diagnosing vascular tumours such as angiosarcomas (Hoffmeier et al., 2014). Statistically, benign tumours are more common than malignant ones. Myxomas are an example of benign tumours and around 50-70% of cardiac tumours are myxomas (Mulisa et al., 2014; Hoffmeier et al., 2014). Moreover, there are different types of malignant tumours including angiosarcomas, which make up 30% of malignant tumours, and rhabdomyosarcoma, which make up 20% of malignant tumours (Hoffmeier et al., 2014). 

Without receiving treatment, the life expectancy is incredibly low at just a few months, and since the best possible treatment is unknown, it is challenging to provide treatment of the highest standard for the patient (Hoffmeier et al., 2014). For example, a simple resection is used for benign tumours, whereas complex tumour resections are used if the tumours are malignant (National Cancer Institute, 2011; Hoffmeier et al., 2014). Artificial heart and heart transplants are used as the last treatment options since these can have complications, revealing the severity of the disease due to the difficulty in treating it and how severe complications can occur, ending in death for most patients (Hoffmeier et al., 2014). 

Even though it is a rare condition, PCS remains a serious issue. However, artificial intelligence (AI) is beginning to be utilised more commonly in different areas of medicine, including in the diagnosis and treatment of cancers. For instance, AI is used in detecting breast cancer due to its high efficiency in reducing false negatives when diagnosing cancer (Lang et al., 2023). By working alongside humans, the detection of breast cancer is more rapid and efficient, since AI may identify areas of concern which some clinicians may miss. Statistics show that AI can detect 20% more breast cancers than humans, which is essential for early and effective treatment (Conner et al., 2024). Furthermore, AI can also personalise treatment plans for patients that target their treatment specifically (Malhotra, 2024). All of these factors result in a more efficient way of treating the patients and an increase in life expectancy. With a disease that is difficult to detect like PCS, the use of AI can change the way in which it is treated, helping save many lives in the future.

This paper will examine the current limitations in diagnosis and treatment of PCS. Furthermore, it will review the AI approaches that are already being used or actively studied. The paper proposes several new AI-driven strategies specifically designed to overcome the lack of data caused by the disease’s rarity. These proposals aim to close critical gaps and shift PCS from an almost certain fatal diagnosis towards one where a meaningful survival rate is achievable.

2. Current Diagnostic Approaches and Limitations

Diagnosing PCS is extremely difficult as it metastasises rapidly and presents nonspecific symptoms (Shanmugam, 2006; Tyebally et al., 2020). Most patients initially feel vague symptoms such as chest discomfort, feeling short of breath or constantly being tired. Consequently, these symptoms are easily mistaken for other issues like heart failure, blood clots or cardiac infections (Shanmugam, 2006; Tyebally et al., 2020). This leads to cancer identification delays, ultimately reducing chances for a complete surgical resection. Surgical resection is the best solution doctors can provide to help patients live longer (Maleki et al., 2017). Another major challenge includes the extreme rarity of PCS, which means that many doctors may never encounter a single case in their careers (Bangolo et al., 2023). This lack of experience makes it common for the tumour to be misdiagnosed as a benign tumour, a blood clot or even pulmonary embolism. These tumours are often misidentified which leads to delayed diagnosis and worsening of the disease (Mulisa et al., 2024; Shanmugam, 2006). Since making a diagnosis early is uncommon, researchers and doctors have been focusing on using several different imaging techniques, including AI, to get a more accurate and faster diagnosis (Burkhard-Meier et al., 2025).

The initial scan doctors perform to observe the heart is usually transthoracic echocardiography (echo). Its frequent use is primarily due to high availability in most hospitals, lower costs and its non-invasive nature. Echo is an effective diagnostic technique that reveals the size of a mass or tissue inside the heart, its movement and any resulting obstruction to blood flow (Mankad & Herrmann, 2016). However, the echo test has some limitations. First, the accuracy of the scan depends on skills obtained by the operator and quality of acoustic windows (Wang et al., 2025). Second, it is difficult for the doctor to characterise tissues, which means it is difficult to diagnose if it is a malignant tumour, benign tumour or a blood clot. In addition, it is also difficult to get a clear picture of any tumours located deep inside the heart or surrounding tissue (Tyebally et al., 2020).

Cardiac magnetic resonance (CMR) imaging is currently the most informative non-invasive way to identify cardiac tumours (Li et al., 2020). It provides highly detailed images that show what the tissue is made of, the definition of the tumour’s edges, the amount of blood supply it receives, the presence of dead tissue (necrosis) and the evidence of metastasis into neighbouring heart muscles. These findings help differentiate sarcomas from benign tumours such as myxomas (Tyebally et al., 2020). However, certain diagnostic challenges remain. For example, some sarcomas may appear like blood clots in the presence of dead tissue or low blood flow, while in other cases harmless tumours can appear aggressive if they have a lot of inflammation (Pazos-López et al., 2014). These confusing overlaps can reduce clinicians’ confidence in their diagnosis, especially at hospitals without experts who regularly deal with heart tumour imaging (Tyebally et al., 2020).

Computed tomography (CT) is another clinically significant tool. A CT scan provides detailed anatomical images showing if the mass has any calcification, if it involves coronary arteries and if the tumour has spread outside the heart (Joudar et al., 2025). This makes CT especially helpful when doctors are planning surgery. However, CT scans do not provide the same high resolution for soft tissue detail as CMR, limiting its ability to reliably differentiate between benign and malignant tumours (Angeli et al., 2024).

Positron emission tomography (PET) using fluorodeoxyglucose (FDG) is essential for staging as PCS often spreads early to lungs, liver and other organs (Isambert et al., 2014; Hammami et al., 2021). Cancerous tumours typically look very bright on a PET scan since they show high FDG indicating high metabolism due to increased sugar in the organs (Rahbar et al., 2012). However, PET scans have limitations. Similar to CMR, PET scans can produce false positives, indicating cancer when it is only an inflammation or swelling of heart muscles or surrounding cardiac tissues. Conversely, it can provide a false negative and miss identifying cancer when tumours are too small or less aggressive (Rahbar et al., 2012; Chang et al., 2006; Shreve et al., 1999). Despite these limitations, PET scans remain a key tool for determining the extent of cancer metastasis (Rahbar et al., 2012; Saponara et al., 2018; Hasani et al., 2021).

Finally, a tissue biopsy provides definitive diagnosis, allowing specialists to examine the cells, tumour markers and molecular features of the growth (Agaimy et al., 2012; Urbini et al., 2020). However, biopsies in cardiac sarcoma are often risky due to the tumour’s fragility, the chance of bleeding and its close proximity to vital heart structures (Isambert et al., 2014; Hasani et al., 2021). Due to these risks, doctors often avoid biopsy and the diagnosis depends entirely on additional imaging tests. Consequently, it is crucial for the scans to be interpreted correctly by the specialists (Saponara et al., 2018; Tyebally et al., 2020).

2.1 Role of AI in Diagnostics

AI is rapidly becoming a promising way to overcome diagnostic limits such as operator dependance, false positives and negatives in assessments, data scarcity and gaps in multimodality imaging (Mushcab et al., 2025; Khera et al., 2024; Ravera et al., 2025). AI-assisted echocardiography can automate image capture, reduce dependence on the operator’s skill and identify structural problems more consistently (Nechita et al., 2025; Mushcab et al., 2025). Specifically, deep learning excels at recognising heart structures and detecting masses, even with varying image quality (Khera et al., 2024; Ravera et al., 2025). Additionally, this support is particularly helpful in hospitals lacking specialised cardiac imaging expertise (Nechita et al., 2025; Ravera et al., 2025).

Radiomics, an AI technique, studies thousands of subtle image features from CMR scans that cannot be detected by clinicians (Mushcab et al., 2025; Yang et al., 2025). These features include the texture patterns, shape details and changes in signal intensity (Yang et al., 2025). Recent research shows that radiomics is highly accurate at helping doctors identify the difference between tumours that are malignant and those that are benign (Mushcab et al., 2025; Khera et al., 2024). Furthermore, deep learning models can learn complex patterns directly from imaging data (Nechita et al., 2025; Yang et al., 2025). Some of these models can even predict the detailed tissue features of a CMR scan by simply analysing basic echocardiography images, opening up new ways to quickly identify patients in need of treatment (Mushcab et al., 2025; Ravera et al., 2025).

Researchers are developing multimodal AI systems that combine information from echocardiography, CMR, CT, PET and patient data to create a single prediction score (Mushcab et al., 2025; Nechita et al., 2025; Yang et al., 2025). This approach mimics how doctors interpret many tests together and can help resolve conflicting imaging results (Khera et al., 2024; Ravera et al., 2025). Since early diagnosis greatly improves the chance of successful tumour removal, AI-supported triage could make a meaningful difference in survival (Mushcab et al., 2025; Nechita et al., 2025; Khera et al., 2024).

2.2 Ethical Challenges of AI Use in Diagnostics

One of the main challenges for AI in PCS diagnosis is the scarcity of high quality data. PCS is so rare that there are few real examples for training AI models and most public cardiac imaging data available is primarily on benign tumours (Sengupta et al., 2024; Germain et al., 2025; Hasani et al., 2021). The lack of real world PCS data limits accuracy in detecting malignant growths and raises fairness concerns as the models may not function effectively across distinct patient groups or hospital settings (Decherchi et al., 2021; Sengupta et al., 2024).

Ethical and governance issues remain a concern. Many advanced AI systems are like black boxes, providing predictions without revealing how specific imaging features influenced the decision. This limited explainability makes doctors less likely to trust the AI when dealing with life-threatening conditions like PCS (Marey et al., 2024; Blackman & Veerapen, 2025). Additionally, regulations for AI in rare diseases are still evolving and concerns regarding data privacy, accountability and fixing bias remain to be addressed (Pham, 2025; Goktas & Grzybowski, 2025). Figure 1 indicates that AI is in a high risk, low readiness zone where black box decisions, bias and unclear regulations still limit clinical trust (Decherchi et al., 2021; Marey et al., 2024).

Figure 1: PCS AI readiness and risk radar (Created by Natasha Suresh, 2025).

2.3 Hybrid Human and AI Diagnostic Model

A hybrid diagnostic model that combines AI with human expertise is the most effective approach for diagnosing rare conditions such as PCS (Sengupta et al., 2024; Mooghali et al., 2024). In this approach, AI helps in detecting subtle abnormalities, processing complex imaging features and synthesising multimodal data from echocardiography, CMR, CT and PET. Additionally, clinicians provide the judgement needed to interpret these findings safely (Khera et al., 2024; Mastrodicasa & van Assen, 2024). This human and AI collaboration addresses risks with using only AI while providing tools to clinicians to ensure diagnostic accuracy (Chen et al., 2024; Quer et al., 2021; Marey et al., 2024).

3. Critical Analysis of Diagnostic Approaches

A deeper analysis of current diagnostic strategies for PCS shows that the main struggles in early detection and accuracy come from three areas that are all connected. These include the biological complexity of PCS, unequal access to specialised care and the technological limitations of both imaging scans and AI (Tyebally et al., 2020; Mulisa et al., 2024; Sengupta et al., 2024). These issues collectively create a situation where delays in diagnosis become increasingly common, causing many patients to lose their chance for a cure (Hammami et al., 2021; Urbini et al., 2020).

The biological complexity of PCS remains one of the main issues. The early signs frequently mimic common heart problems, causing doctors to treat for heart failure, myocarditis or blood clots instead of suspecting a tumour (Shanmugam, 2006; Mulisa et al., 2024). Despite the fact that advanced scanning techniques are being used, features of the tumour like necrosis, inflammation and irregular edges frequently look similar for both malignant and benign tumours. Consequently, this limits the accuracy of advanced scans like CMR and PET, reducing confidence in diagnosis (Rahbar et al., 2012; Pazos-López et al., 2014; Tyebally et al., 2020).

AI offers new possibilities but also introduces major challenges. AI can analyse subtle imaging patterns, combine information from multiple modals and reduce human errors. However, the success of AI depends on large, varied and structured datasets, which do not currently exist for PCS. The current AI models are primarily trained on common, benign tumours. Hence, their ability to correctly recognise malignant patterns in rare tumours like PCS is limited (Decherchi et al., 2021; Hasani et al., 2021; Sengupta et al., 2024). This increases the risk of AI misclassifying a tumour and reduces its accuracy. A second major concern is the lack of external validation. Most AI models are tested only on the data they were trained on or data from the same hospital. When these models are used in new hospitals or with different patient groups, their accuracy reduces. Without studies done across many different hospitals and multi-centre validation, the AI tools will not work consistently for patients (Germain et al., 2025; Hasani et al., 2021).

Explainability is also a major ethical problem. Many AI systems produce results that are difficult to interpret, making it difficult for clinicians to understand the reasoning behind certain predictions.This lack of clarity can make doctors hesitate to rely on AI’s assessment of whether a tumour is malignant or not (Marey et al., 2024; Blackman & Veerapen, 2025). Recent methods, like heat maps and confidence indicators, are being developed to increase the transparency of AI decision making. However, these tools are not used everywhere. There are also concerns about data governance. Since PCS is rare, sharing data among different hospitals is essential for training reliable AI models. However, privacy laws make this data sharing difficult. A technique called federated learning is becoming a potential solution, allowing hospitals to work collaboratively to train AI models without actually sharing the patient data itself (Rieke et al., 2020; Sheller et al., 2020). This approach may be especially important for studying PCS.

Structural inequities due to unfair differences in the healthcare system worsen diagnostic outcomes. Since PCS is so rare, only a handful of hospitals globally have the necessary expertise to read the complex imaging scans correctly. Additionally, patients in rural and underserved regions face delays in getting a referral, receiving scans and getting a diagnosis from a specialist. This means that receiving a diagnosis is heavily influenced by where a person lives and which resources are available in their area, rather than the patient’s specific needs (Ohman et al., 2021; Foster et al., 2024). These unfair differences contribute to poor survival outcomes.

Finally, AI systems can be overconfident, giving predictions with high certainty even when wrong. Without proper safeguards, this could easily mislead clinicians or strengthen any existing bias (Chan, 2023; Marey et al., 2024).

In summary, AI has significant strengths but cannot independently resolve the challenges of diagnosing PCS by itself. Consequently, the most realistic and responsible way forward is a hybrid model where AI supports clinicians by analysing complicated imaging data, while human experts provide the contextual interpretation, judgement and oversight (Khera et al., 2024; Mooghali et al., 2024). This combined approach deals with the limitations of both human and machine decision making and offers the safest way to improve the early diagnosis of this rare and aggressive cancer.

4. Current Treatment Modalities and Limitations

PCS is not only uncommon, it is also surgically and medically one of the most dangerous malignancies within oncology. Although numerous histological subtypes exist, PCS is clinically categorised based on anatomical location into right heart and left heart sarcoma (Cao et al., 2025; Burkhard-Meier et al., 2025; Vaporciyan & Reardon, 2010). Its rarity results in fewer than one in 500 open-heart operations, ultimately limiting the range of treatment modalities available (Tyebally et al., 2020).

Right heart sarcoma extends to the exterior of the heart; its physical properties make it bulky and infiltrative. Additionally, it has the ability to metastasise early in the process (Aviel et al., 2019; Guan et al., 2022). Survival rate for right heart sarcoma if untreated is dismal, yet right heart sarcoma rarely results in heart failure unless the disease significantly progresses (Vaporciyan & Reardon, 2010). The primary treatment for right heart sarcoma includes neoadjuvant chemotherapy. The aim of the pre-surgery neoadjuvant chemotherapy is to reduce tumour volume, ultimately increasing prospects for a successful R0 resection – a complete surgical removal of the tumour with microscopical negative margins, lacking residual tumour. Depending on the tumour’s response to the neoadjuvant chemotherapy, surgery might be performed after the fourth or sixth round of the therapy (Vaporciyan & Reardon, 2010). 

In contrast to right heart sarcoma, left heart sarcoma is more solid and less infiltrative. Even though metastasis occurs later in progress compared to right heart sarcoma, there is an increased risk of heart failure and treatment is more limited. In other words, due to higher frequency of heart failure, treatment options for right heart sarcoma and left heart sarcoma are dissimilar (Blackmon & Reardon, 2025c). Due to the elevated risks for heart failure, neoadjuvant chemotherapy is commonly not an option. The suitable approach would instead be surgical excision in order to avoid heart failure. However, due to the anatomic location of the left heart sarcoma, it is a difficult surgical operation and R0 might not be achieved (Chan et al., 2022). Most common outcomes for a left heart sarcoma surgery is R1 resection (a microscopical tumour which is not visible to the naked eye but can be observed using a microscope) or a R2 resection with visible tumour remaining after the surgery (Blackmon & Reardon, 2025c; Mulisa et al., 2024).

4.1 Limitations in Treatment and Clinical Decision Making

Due to the rarity of PCS, therapeutic options are limited and few surgeons have accumulated the required expertise to efficiently treat the tumour (Blackmon & Reardon, 2025b). As previously mentioned, primary cardiac tumour incident rates are extremely low: approximately one in every 500 open heart surgeries are related to primary cardiac sarcoma; 25% of heart tumours are malignant; and 75% of the malignant tumours are primary cardiac sarcomas (Blackmon & Reardon, 2025b). Consequently, the lack of clinical experience increases reliance on case studies, retrospective data and small case series rather than prospective data, which ultimately creates a challenge for standardised management of primary cardiac sarcoma. 

Additionally, complications arise from the anatomic heterogeneity of primary cardiac sarcoma. The right heart sarcoma is often found in the right atrium. Its infiltrative growth patterns and early metastasising makes complete surgical resection extremely difficult (Vaporciyan & Reardon, 2010; Aviel et al., 2019). Moreover, the left heart sarcoma imposes significant risks for heart failure, excluding the possibility for neoadjuvant chemotherapy to reduce tumour volume prior to surgery, R0 becomes an impossible goal (Blackmon & Reardon, 2025b). 

These histological and anatomic varieties create prognostic uncertainty and further complicate surgical decision making. The chances of achieving a curative surgery are severely reduced as a result of the infiltrative nature of right heart sarcomas and the elevated risks for recurrence in the left heart lesions (Cao et al., 2025). Additionally, even in the case of successful R0, metastasis and recurrence is common, as reported in retrospective studies.  

Ultimately, the prognosis for primary sarcoma is poor, and most patients experience rapid metastasis and recurrence. According to Blackmon and Reardon (2025b), 90% of primary cardiac sarcoma patients have deceased 6-12 months after initial tumour growth. In cases where R0 is achieved, long term predictions of outcomes remain a challenge due to limited case studies and diversity of tumour behaviour.

4.2 Role of AI in Treatment Planning

AI is an emerging tool used for diagnostics and managing rare malignancies such as PCS. While clinical trials for AI use remains in early stages, AI driven approaches have been shown to provide increased diagnostic accuracy and prognostic assessment, as well as provide suitable treatment plans for individual patients (Amandine Crombé et al., 2023). Cardiac radiologists can be assisted by AI algorithms for interpretations of CT imaging and cardiac MRI imaging, allowing higher tumour distillation and acknowledgement of critical structures otherwise overlooked by radiologists (Angeli et al., 2024; Li et al., 2020). Discoveries made by AI can consequently be utilised to enhance treatment planning and preoperative assessment (Ravera et al., 2025; Hasani et al., 2021). 

Moreover, AI based predictive models are able to include integrated multimodal data allowing more holistic overviews and aiding with surgical planning, prognostic predictions and risk stratification (Nechita et al., 2025; Ravera et al., 2025). The integrated multimodal data considers histologic pathology, imaging, genomic profiling and clinical variables. Within pathology, AI has shown promise in detection of microscopical invasion and facilitated reproductive grading, allowing characterisation of sarcoma subtype. This is particularly useful for diagnosing and treating the complexity of primary cardiac sarcomas.

Although no use of AI is currently validated specifically for PCS, general development in radiomics, digital pathology and algorithms for decision support suggests future use of AI for this rare disease (Mushcab et al., 2025; Yang et al., 2025). Integration of AI tools and radiologists can help compensate for the limited clinical experience, as well as enhance individualised and evidence-based treatment management (Ravera et al., 2025; Blackman & Veerapen, 2025).

5. Critical Analysis of Treatment Strategies

A critical examination of the current therapeutic strategies for PCS shows that successful management is challenged by three key challenges. They include a lack of a standard evidence-based treatment plan, extreme anatomical and biological constraints limiting curative surgery, and ethical and data related challenges for using AI to deliver treatments (Tyebally et al., 2020; Burkhard-Meier et al., 2025; Chan et al., 2022). Consequently, these limitations result in reactive care for PCS, focused on symptom management and comfort rather than curative treatment. This directly leads to poor survival outcomes for most patients (Blackmon & Reardon, 2009).

A key problem is the absence of standardised, PCS-specific guidelines due to the tumour’s extreme rarity. Due to this data deficit, the clinicians rely on individual experiences, case reports and small past studies instead of the large clinical trials that could create standards and guidelines. As a result, treatment decisions remain inconsistent, including differences in the use of chemotherapy administered before (neoadjuvant) and after (adjuvant) surgery. Most existing guidelines are derived from guidelines designed for sarcomas in other parts of the body (Burkhard-Meier et al., 2025). This absence of clear, agreed-upon rules compromises the ability to reliably control cancer cells that have spread subtly, which is the main reason for early recurrence and metastatic spread, even after surgery.

Although the goal is R0 resection (complete tumour removal), success rates remain low due to the heart’s critical location. Right heart sarcomas are bulky, making R0 nearly impossible while the left heart sarcomas have the risk of acute heart failure, often preventing the necessary chemotherapy required prior to surgery (Vaporciyan & Reardon, 2010; Blackmon & Reardon, 2009). Surgeons must balance preservation of cardiac function with achieving complete tumour removal and adequate surgical margins. This often leads to leaving cancer cells behind (R1 or R2 disease), which is the direct cause of poor prognosis (Guan et al., 2022).

Training AI to predict patient outcomes such as successful R0 resection or chemotherapy response requires substantial, integrated datasets of long-term results that currently do not exist for PCS (Hasani et al., 2021). This data gap limits AI’s current use in critical surgical planning. Furthermore, the use of AI deployment raises accountability issues and if a treatment suggested by AI fails, the line of responsibility is unclear (Goktas & Grzybowski, 2025). This, combined with the lack of transparency in AI decision making (black box) creates resistance amongst clinicians to use it.

These challenges highlight why a hybrid approach of human and AI remains essential in exploring future treatment strategies for PCS (Khera et al., 2024).

6. Proposal: Leveraging AI to Close Diagnostic and Treatment Gaps

To reduce the rate of PCS, theis proposal harnesses the transformative potential of AI to address the critical gaps in diagnosing and treating PCS. 

Due to its rarity, PCS presents non-specific symptoms such as fatigue, chest discomfort or embolic events that frequently lead to misdiagnosis or delayed detection until advanced stages. Traditional diagnostic tools like echocardiography, MRI and CT scans remain invaluable but limited by reliance on human interpretation, which can miss subtle tumour features or misclassify benign lesions. Deep learning algorithms, such as convolutional neural networks (CNNs), can analyse thousands of imaging datasets to detect minute abnormalities and malignancy indicative patterns that escape human vision, thus enabling earlier and more accurate diagnosis. Integrating AI with advanced imaging modalities further allows for automated segmentation and quantification of tumour size, location and invasion depth, assisting surgeons in planning complex resections. 

Beyond imaging, molecular profiling through techniques like genomics and transcriptomics provides insight into the genetic alterations driving PCS, such as mutations in tumour suppressor genes or aberrant angiogenic pathways. AI models can analyse large-scale molecular data to identify biomarkers predictive of tumour behaviour, prognosis and response to targeted therapies. By combining imaging features with molecular data, AI can classify tumour types with higher precision, facilitating personalised medicine approaches and tailoring treatments based on individual tumour biology rather than a one-size-fits-all strategy. These personalised models could suggest optimal surgical approaches, chemotherapeutic regimens or immunotherapy options, thus improving survival and quality of life, rather than waiting until it is too late. 

Here are the proposals:

PROPOSAL 1

Implementing this approach requires creating a comprehensive, multicentre database that aggregates anonymised imaging, histopathological and molecular data from PCS patients worldwide. Such a database would enable machine learning models to be trained and validated across diverse populations and tumour subtypes, improving generalisability and robustness. Continuous refinement through feedback loops and real-world testing will enhance model accuracy over time. In addition, the deployment of AI tools in clinical workflows promises to reduce diagnostic delays, improve accuracy and support clinicians in making data-driven, personalised treatment decisions. 

PROPOSAL 2

Creating a partnership that brings together major technology companies, such as Google, Apple and Samsung, with healthcare providers to launch a free monitoring system on smartwatches and smartphones. Inspired by research showing that changes in voice can help detect heart disease early, this system would combine passive health tracking like heart rate, heart rhythms and movement patterns with a quick weekly voice check, such as asking users to count to 20 (Hartley & Khamis, 2022). AI models trained on large heart disease datasets and updated using secure learning methods across millions of devices could identify early warning signs months before symptoms appear. When a possible risk is detected, the system would alert the user and their doctor so that follow up heart imaging tests can be ordered for early diagnosis and treatment.

PROPOSAL 3

Introduce an AI powered ‘Virtual Heart Twin’ system to help doctors plan and perform more precise surgeries for PCS. Medical images such as echocardiograms, cardiac MRI, PET and CT scans are used to create a detailed 3D digital model of each patient’s heart and tumour (Shen et al., 2025). Surgeons can use virtual reality to practise complex procedures before surgery and predict crucial outcomes, such as cardiac functions and complication risks. During surgery, virtual reality displays can guide surgeons by highlighting tumour boundaries in real time (Venkatesan et al., 2021). As more cases from around the world are added to the system, the AI will continue to learn and improve. This technology could make heart tumour surgery safer, more accurate and more effective and improve patient survival.

Ultimately, this integration of AI has the potential to shift PCS management from a predominantly palliative approach to one centred on early detection, precision therapy and improved survival outcomes, turning a once nearly incurable disease into a manageable condition.

7. Conclusion

PCS is one of the most difficult cancers to diagnose. It is rare and fatal, and doctors rely on tools such as echocardiography, CT scans, MRIs and occasionally biopsies to diagnose it. However, these methods have limitations. Echocardiograms can miss minor or differently placed tumours. CT and MRI scans provide good detail, yet radiologists may miss key features of a malignant tumour. Moreover, biopsies can be risky due to the tumour’s formation inside the heart, and the tissue sample might not represent the entire tumour. Another factor contributing to the possible late diagnosis of the disease is its broad, nonspecific symptoms, such as weight loss and fatigue.

However, the emerging innovation and rising popularity of AI in the medical field (due to its ability to analyse medical images much more thoroughly than a person can) have growing significance. AI can also evaluate thousands of past cases, detect patterns and compare current images to a bulk of datasets. This can help detect a tumour earlier, flagging abnormalities quickly and providing doctors with more information like subtle abnormalities between comparisons and quantitative measurements. AI has the potential to help in this area due to its ability to look at genetic profiles, treatment outcomes, tumour behaviour and patient health factors. These are all essential factors that can help predict which treatment plans might work best for patients. For example, AI models can estimate how likely a tumour is to respond to chemotherapy or how risky a surgery might be based on the tumour’s size and position. It can also help plan surgeries by showing surgeons detailed 3D maps of the tumour and predicting how removing certain parts might affect heart function. Although the difficulty of the treatment remains the same, AI can boost efficiency and precision of diagnostics and treatment.

AI can be used constructively, however it also has significant limitations. AI models only work as well as the data they are trained on, and because PCS is so rare, there is not enough data to cover every case. This can make AI predictions inaccurate. There are also ethical issues, like protecting patient privacy, avoiding bias and knowing how much to trust AI when making decisions. It’s also hard to assign responsibility if AI gives an incorrect recommendation that leads to harm. Another problem is AI’s lack of empathy and consideration of patient history. Therefore,  the best approach is the integration of humans and AI. Doctors can make final decisions, interpret patient context and act ethically, while AI can quickly analyse lots of information and spot patterns humans might miss. In other words, they can collaboratively work to ensure best possible patient care. Despite current AI flaws, it demonstrates potential. Traditional methods are reliable, however their efficiency decreased when used for rare cancers such as PCS. As technology improves and more data becomes available, AI could help create personalised treatment plans that adapt to individual patient response.

Overall, PCS is one example that displays the importance of collaboration between human judgement and advanced technology within oncology. Traditional diagnostic and treatment methods provide experience and reliability. By implementing the usage of AI, which can bring speed, precision and the analysis of patterns that humans cannot see, the medical field can create a more ethical and efficient approach. Although there are still obstacles like data limitations, ethical questions and the need for better AI transparency, the future holds hopeful opportunities. With continued research and a deeper understanding of both the disease and the potential of artificial intelligence, how we diagnose and treat primary cardiac sarcoma can be transformed. This gives patients a much better chance at earlier detection, clearer decisions and more personalised care. To fully realise this potential, researchers, clinicians and policymakers must collaborate to improve data quality, establish ethical AI standards and implement transparent, accountable AI systems across oncology, epidemiology and broader medical practice.

Agaimy, A., Rösch, J., Weyand, M. and Strecker, T. (2012). Primary and metastatic cardiac sarcomas: a 12-year experience at a German heart center, International Journal of Clinical and Experimental Pathology, 5(9), pp. 928–938. 

Amandine. C, Spinnato, P., Italiano, A., Brisse, H.J., Feydy, A., Fadli, D. and Kind, M. (2023). Radiomics and artificial intelligence for soft-tissue sarcomas: Current status and perspectives, Diagnostic and Interventional Imaging, 104(12), pp. 567–583. 

Angeli, F., Bodega, F., Bergamaschi, L., Armillotta, M., Amicone, S., Canton, L., Fedele, D., Suma, N., Cavallo, D., Foà, A., Belmonte, M., Russo, V., Attinà, D., Niro, F., Bonfiglioli, R., Fanti, S., Pavon, A.G., Guglielmo, M., Mushtaq, S. and Pantaleo, M.A. (2024). Multimodality Imaging in the Diagnostic Work-Up of Patients With Cardiac Masses, JACC: CardioOncology, 6(6), pp. 847–862.

Aviel, G., Meirovitz, A., Planer, D., Cohen, D., Gilon, D., Feldman, J. and Shapira, O.M. (2019). Endovascular brachytherapy for extensive right-heart and pulmonary artery sarcoma – a case report,  Journal of Contemporary Brachytherapy, 11(6), pp. 579–583.

Ayrton. B, Fwelo, P., Iyer, K.M., Klinger, S., Benathar, L., Dey, S., Angel Ann Chacko, Hein, M., Samyukta Gudena, Lawal, G., Sivasubramanian, B.P., Zekordavar Rimba, Kinjal Hirpara, Merajunnissa Merajunnissa, Swathi Veliginti, Arana, G., Sathyarajan, D.T., Sachin Kumar Singh, Shetty, T. and Bhardwaj, K. (2023). Primary Cardiac Sarcoma: Clinical Characteristics and Prognostic Factors over the Past 2 Decades, Diseases, 11(2), pp. 74–74.

BCRF (2024). AI and Personalized Breast Cancer Care, Breast Cancer Research Foundation [online]. <https://www.bcrf.org/blog/ai-predictive-analytics-precision-medicine-breast-cancer/>

Blackman, J. and Veerapen, R. (2025). On the practical, ethical, and legal necessity of clinical artificial intelligence explainability, BMC Medical Informatics and Decision Making, 25(1), p. 111. 

Blackmon, S.H. and Reardon, M.J. (2009). Surgical Treatment of Primary Cardiac Sarcomas, Texas Heart Institute Journal, 36(5), p.451. 

Burkhard-Meier, A., Gioia, D.D., Vindi Jurinovic, Hoberger, M., Güler, S.E., Völkl, M., Corradini, S., Gaasch, A.V., Alig, A.H.S., Knösel, T., Hagl, C., Schneider, C., Wulf Sienel, Kunz, W.G., Caspar Burkhard-Meier, Bergwelt-Baildon, M. von, Lindner, L.H. and Berclaz, L.M. (2025). Primary cardiac sarcoma: insights from two decades of multimodal management at LMU Munich, Cardio-Oncology, 11(1), pp. 58–58.

Chan, E.Y., Ali, A., Zubair, M.M., Nguyen, D.T., Ibarra-Cortez, S.H., Graviss, E.A., Shapira, O.M., Ravi, V., MacGillivray, T.E. and Reardon, M.J. (2022). Primary cardiac sarcomas: Treatment strategies, Journal of Thoracic and Cardiovascular Surgery, 166(3), pp. 828–838.

Chan, B. (2023). Black-box assisted medical decisions: AI power vs. ethical physician care, Medicine Health Care and Philosophy, 26(3), pp. 285–292. 

Chang, J.M., Lee, H.J., Goo, J.M., Lee, H.Y., Lee, J.J., Chung, J.K. and Im, J.G. (2006). False positive and false negative FDG-PET scans in various thoracic diseases, Korean Journal of Radiology, 7(1), pp. 57–69. 

Chauhan, A., Agarwal, A., Naqvi, H. and Geelani, M.A. (2018). Malignant Tumors of Heart: A Review, World Journal of Cardiovascular Surgery, 8(11), pp. 204–212. 

Chen, M., Wang, Y., Wang, Q., Shi, J., Wang, H., Ye, Z., Xue, P. and Qiao, Y. (2024). Impact of human and artificial intelligence collaboration on workload reduction in medical image interpretation, npj Digital Medicine, 7(349), pp. 1–10.

Conner, K. (2024). Using AI (Artificial Intelligence) to Detect Breast Cancer, Breastcancer.org [online]. <https://www.breastcancer.org/screening-testing/artificial-intelligence>

Decherchi, S., Pedrini, E., Mordenti, M., Cavalli, A. and Sangiorgi, L. (2021). Opportunities and challenges for machine learning in rare diseases, Frontiers in Medicine, 8, p. 747612.

Ding, D.-C., Shyu, W.-C. and Lin, S.-Z. (2011). Mesenchymal Stem Cells, Cell Transplantation, 20(1), pp. 5–14. 

Foster Osei Baah, Sharda, S., Davidow, K., Jackson, S., Kernizan, D., Jacobs, J.A., Baumer, Y., Schultz, C.L., Baker-Smith, C.M. and Powell-Wiley, T.M. (2024). Social Determinants of Health in Cardio-Oncology, JACC. CardioOncology, 6(3), pp. 331–346.

Germain, D.P., Gruson, D., Malcles, M. and Garcelon, N. (2025). Applying artificial intelligence to rare diseases: a literature review highlighting lessons from Fabry disease, Orphanet Journal of Rare Diseases, 20(1), p. 186. 

Goktas, P. and Grzybowski, A. (2025). Shaping the future of healthcare: ethical clinical challenges and pathways to trustworthy AI, Journal of Clinical Medicine, 14(5), p. 1605. 

Guan, T., Wei, Q., Tang, Y., Zhao, H., Lu, Z., Feng, W., Teng, Y., Luo, Z., Chi, K., Ou, C. and Chen, M. (2022). Metastatic patterns and prognosis of patients with primary malignant cardiac tumor, Frontiers in Cardiovascular Medicine, 9(1009765), pp. 1–10.

Hammami, M.B., Al-Wawi, M.Z., Fazel, H. and Oudih, M. (2021). Incidence, prognostic significance, and survival outcomes of primary cardiac sarcoma: an updated population-based retrospective study, European Research Journal of Medical Investigation, 11(2), pp. 45–52.

Hartley, A. and Khamis, R. (2022). Voice Biomarkers: The Most Modern and Least Invasive Tool for Coronary Assessment?, Mayo Clinic Proceedings, 97(5), pp. 816-818. 

Hasani, N., Farhadi, F., Morris, M.A. et al. (2021). Artificial intelligence in medical imaging and its impact on the rare disease community: Threats, Challenges and Opportunities, PET Clinics, 17(1), pp. 13–29. 

Hoffmeier, A., Sindermann, J.R., Scheld, H.H. and Martens, S. (2014). Cardiac Tumors: Diagnosis and Surgical Treatment, Deutsches Aerzteblatt Online, 111(12), pp. 205–211.

Isambert, N., Ray-Coquard, I., Italiano, A., Rios, M., Kerbrat, P., Gauthier, M., Blouet, A., Chaigneau, L., Duffaud, F., Piperno-Neumann, S., Kurtz, J.-E., Girard, N., Collard, O., Bompas, E., Penel, N., Bay, J.-O., Guillemet, C., Collin, F., Blay, J.-Y. and Le Cesne, A. (2014). Primary cardiac sarcomas: A retrospective study of the French Sarcoma Group, European Journal of Cancer, 50(1), pp.128–136. 

Joudar, I., Nasri, S., Kamaoui, I. and Skiker, I. (2025). Cardiac computed tomography for cardiac masses: a necessity or a luxury?, Journal of International Medical Research, 53(2), pp. 1–13.

Khera, R., Asnani, A.H., Krive, J., Addison D. et al. (2024). Artificial intelligence to enhance precision medicine in cardio-oncology: A Scientific Statement From the American Heart Association, Circulation: Genomic and Precision Medicine, 17(5), p. e0000097.

Lång, K., Josefsson, V., Larsson, A.-M., Larsson, S., Högberg, C., Sartor, H., Hofvind, S., Andersson, I. and Rosso, A. (2023). Artificial intelligence-supported screen reading versus standard double reading in the Mammography Screening with Artificial Intelligence trial (MASAI): a clinical safety analysis of a randomised, controlled, non-inferiority, single-blinded, screening accuracy study, Lancet Oncology, 24(8), pp. 936–944. 

Li, X., Chen, Y., Liu, J., Xu, L., Li, Y., Liu, D., Sun, Z. and Wen, Z. (2020). Cardiac magnetic resonance imaging of primary cardiac tumors, Quantitative imaging in medicine and surgery, 10(1), pp. 294–313. 

Maleki, M.H., Moein Aboobakri Makouei, Hatami, F. and Reza Zeinabadi Noghabi (2017). Primary Cardiac Synovial Sarcoma: A Case Report, The Journal of Tehran University Heart Center, 12(1), p. 32.

Mankad, R. and Herrmann, J. (2016). Cardiac tumors: echo assessment, Echo Research & Practice, 3(4), pp. R65–R77. 

Marey, A., Arjmand, P., Alerab, A.D.S. et al. (2024). Explainability, transparency and black box challenges of AI in radiology: impact on patient care in cardiovascular radiology, Egyptian Journal of Radiology and Nuclear Medicine, 55(1), p. 183. 

Mastrodicasa, D. and van Assen, M. (2024). Artificial intelligence for cardiac imaging is ready for widespread clinical use: Pro Con debate AI for cardiac imaging, BIR Open, 7(1), pp. 1–11.

Moeri-Schimmel, R., Pras, E., Desar, I., Krol, S. and Braam, P. (2020). Primary sarcoma of the heart: case report and literature review,  Journal of Cardiothoracic Surgery, 15(1), pp. 1–6.

Mooghali, M., Stroud, A.M., Dong Whi Yoo, Barry, B.A., Grimshaw, A.A., Ross, J.S., Zhu, X. and Miller, J.E. (2024). Trustworthy and ethical AI-enabled cardiovascular care: a rapid review, BMC Medical Informatics and Decision Making, 24(1), pp. 1–12.

Mushcab, H., Al Ramis, M., Al Rujaib, A. et al. (2025). Application of artificial intelligence in cardio-oncology imaging, JMIR Cancer, 11(1), p. e63964. 

Mulisa, M.D., Bekele, S.K., Woldemichael, S.B., Gaito, S.G., Abrar, F.N. and Hailu, B.G. (2024). Primary Cardiac Sarcoma, Diagnostic and Therapeutic Challenges in Resource Limited Setting: Case Report, International medical case reports journal, 17, pp. 57–61. 

National Cancer Institute (2011). NCI Dictionary of Cancer Terms, National Cancer Institute [online]. <https://www.cancer.gov/publications/dictionaries/cancer-terms/def/resection>

National Heart, Lung, and Blood Institute (2022). Cardiomyopathy, NHLBI, NIH [online]. <https://www.nhlbi.nih.gov/health/cardiomyopathy>

Nechita, L.C., Tutunaru, D., Nechita, A. et al. (2025). AI and smart devices in cardio-oncology: Advancements in Cardiotoxicity Prediction and Cardiovascular Monitoring, Diagnostics, 15(6), p. 787. 

Ohman, R., Yang, E.H. and Abel, M.L. (2021). Inequity in Cardio‐Oncology: Identifying Disparities in Cardiotoxicity and Links to Cardiac and Cancer Outcomes, Journal of the American Heart Association, 10(24), pp. 1–18.

Pazos-López, P., Pozo, E., Siqueira, M.E., García-Lunar, I., Cham, M., Jacobi, A., Macaluso, F., Fuster, V., Narula, J. and Sanz, J. (2014). Value of CMR for the Differential Diagnosis of Cardiac Masses, JACC: Cardiovascular Imaging, 7(9), pp. 896–905. 

Pham, T. (2025). Ethical and legal considerations in healthcare AI: innovation and policy for safe and fair use, Royal Society Open Science, 12(5), pp. 1–24.

Quer, G., Arnaout, R., Henne, M. and Arnaout, R. (2021). Machine Learning and the Future of Cardiovascular Care, Journal of the American College of Cardiology, 77(3), pp. 300–313.

Rahbar, K., Stegger, L., Schäfer, J. et al. (2012). Differentiation of malignant and benign cardiac tumors using 18F-FDG PET/CT, Journal of Nuclear Medicine, 53(6), pp. 856–863.

Ravera, V., Gilardi, N., Ballestrero, A. and Zoppoli, G. (2025). Applications, challenges and future directions of artificial intelligence in cardio-oncology, European Journal of Clinical Investigation, 55(Suppl 1), p. e14370. 

Rieke, N., Hancox, J., Li, W., Milletarì, F., Roth, H.R., Albarqouni, S., Bakas, S., Galtier, M.N., Landman, B.A., Maier-Hein, K., Ourselin, S., Sheller, M., Summers, R.M., Trask, A., Xu, D., Baust, M. and Cardoso, M.J. (2020). The future of digital health with federated learning, npj Digital Medicine, 3(1), pp. 1–7. 

Saponara, M., Indio, V., Pizzi, C. et al. (2018). 18F-FDG-PET/CT imaging in cardiac tumors: illustrative clinical cases and review of the literature, Nuclear Medicine and Molecular Imaging, 52(5), pp. 358–366. 

Sengupta, P.P., Dey, D., Davies, R.H., Duchateau, N. and Yanamala, N. (2024). Challenges for augmenting intelligence in cardiac imaging, The Lancet Digital Health, 6(10), pp. e739–e748. 

Shanmugam, G. (2006). Primary cardiac sarcoma, European Journal of Cardio-Thoracic Surgery, 29(6), pp. 925–930. 

Sheller, M.J., Edwards, B., Reina, G.A., Martin, J., Pati, S., Kotrotsou, A., Milchenko, M., Xu, W., Marcus, D., Colen, R.R. and Bakas, S. (2020). Federated learning in medicine: facilitating multi-institutional collaborations without sharing patient data, Scientific Reports, 10(1), p. 12598.

Shen, S., Qi, W., Liu, X., Zeng, J., Li, S., Zhu, X., Dong, C., Wang, B., Shi, Y., Yao, J., Wang, B., Jing, L., Cao, S. and Liang, G. (2025). From virtual to reality: innovative practices of digital twins in tumor therapy, Journal of Translational Medicine, 23(1), pp. 1–31.

Shreve, P.D., Anzai, Y. and Wahl, R.L. (1999). Pitfalls in oncologic diagnosis with FDG PET imaging, RadioGraphics, 19(1), pp. S47–S64. 

Sun, Y.-P., Wang, X., Gao, Y.-S., Zhao, S. and Bai, Y. (2017). Primary cardiac sarcoma complicated with cerebral infarction and brain metastasis: A case report and literature review, Cancer Biomarkers, 21(1), pp. 247–250.

Tyebally, S., Chen, D., Bhattacharyya, S. et al. (2020). Cardiac tumors: JACC CardioOncology state-of-the-art review, JACC: CardioOncology, 2(2), pp. 293–311.

Ugarte, S., Sandhu, R.S., Sung, J. and Knopf, K.B. (2021). Primary undifferentiated pleomorphic cardiac sarcoma presenting as right heart failure, Radiology Case Reports, 16(4), pp.829–834.

Urbini, M., Astolfi, A., Indio, V. et al. (2020). Genetic aberrations and molecular biology of cardiac sarcoma, Therapeutic Advances in Medical Oncology, 12, p. 1758835920918492.

Vaporciyan, A. and Reardon, M.J. (2010). Right Heart Sarcomas, Methodist DeBakey Cardiovascular Journal, 6(3), pp. 44–48. 

Venkatesan, M., Mohan, H., Ryan, J.R., Schürch, C.M., Nolan, G.P., Frakes, D.H. and Coskun, A.F. (2021). Virtual and augmented reality for biomedical applications, Cell Reports Medicine, 2(7), p.100348. 

Wang, C., Li, X., Yang, L. and Zhang, W. (2025). Cardiac tumors: clinical presentation, diagnosis, treatment, and results, Langenbeck’s Archives of Surgery, 410(1), p. 228.

Xu, W., Hao, D., Hou, F., Zhang, D. and Wang, H. (2020). Soft Tissue Sarcoma: Preoperative MRI-Based Radiomics and Machine Learning May Be Accurate Predictors of Histopathologic Grade, American Journal of Roentgenology, 215(4), pp. 963–969. 

Yang, J., Ye, L. and Feng, Z. (2025). Application of artificial intelligence in medical imaging: current status and future directions, iRadiology, 3(1), p. 70008.