Glioma and other primary CNS neoplasms

Purpose

Estimation the histopathological type, grade, gene mutation and outcome of gliomas and other primary CNS neoplasms from MRI

Tag(s)

 

Panel

Neuroradiology

Define-AI ID

19020014

Originator

Houman Sotoudeh, MD
Lead Houman Sotoudeh, MD

Panel Chair

Alex Norbash, MD

Panel Reviewers

Neuroradiology Panel

License

Creative Commons 4.0 
Status Public Commenting
RadElement Set RDES115
                               

Clinical Implementation


Value Proposition

Primary CNS tumors are challenging conditions with glioma being the most common primary CNS neoplasm. The definite diagnosis of these primary CNS malignancies necessitates biopsy, pathological evaluation and gene sequencing. At this time accurate prediction of histopathology type, grading and especially gene mutation-outcome is not possible via evaluation of MR images. The role of “radiomics” is evolving and AI models are promising in this field. If an AI algorithm can be trained on the brain MRIs to predict the histopathology, grade, gene mutation and prognosis of glioma, the neurosurgeon and patients can be prepared for different treatment plans.

Narrative(s)

An otherwise healthy 30 year old man presents with recent onset seizure. The brain MRI shows an ill-defined T2/FLAIR hyper-signal intensity in anterior right temporal lobe with extension to the right insula without abnormal enhancement or increased rCBV on MR perfusion. Images are most consistent with low grade glioma but for the definite diagnosis the patient needs biopsy.

Workflow Description

Collecting the pre-op/pre-biopsy brain MRI in different sequences in patients with glioma and other primary CNS neoplasms. Collecting the pathological diagnosis, grade and gene mutation in each patient as well as survival rate, presence or absence pseudo-progression, pseudo-response, recurrence and radiation necrosis after surgery, chemo and radiation.

The image is obtained from MR scanner and sent to PACS and the AI engine. Images are analyzed by the engine. The system then detects and predicts the possible histopathology, grade, gene mutation and prognosis of glioma and other primary CNS neoplasms as well as a prediction about chance of developing radiation necrosis, pseudoprogression and psudo-response after treatments and finally estimated survival rate . An alert message is sent to PACS from the engine with the information and graphic highlighting these details.

Considerations for Dataset Development


The algorithm must be trained on all pre-op MRI sequences from patients with primary CNS neoplasm who are candidates for biopsy or surgical resection. The algorithm also must be trained with attention to the result of post biopsy/surgical histopathology report including the tumor type, grade, gene mutation as well as medical conditions after treatment including presence or absence of recurrence, radiation necrosis, pseudoresponse, pseudoprogression and finally the survival rate.

Technical Specifications


Inputs

DICOM Study

Procedure

Mr, Brain

Views

All

Data Type

DICOM

Modality

MR

Body Region

Head

Anatomic Focus

Brain

Pharmaceutical

N/A

Scenario

Before biopsy and surgery


Primary Outputs


Tumor Classification

RadElement ID

RDE746

Definition

Classify glioma and other primary CNS neoplasms

Data Type

Categorical

Value Set

  • Astrocytoma
  • Oligdendroglioma
  • Oligastrocytoma
  • Ependymoma
  • Ganglioglioma
  • Other primary CNS tumor given the WHO classification

Units

N/A

Tumor Grade

RadElement ID

RDE747

Definition

Tumor grade

Data Type

Categorical

Value Set

  • Grade I
  • Grade II
  • Grade III
  • Grade IV

Units

N/A


Identify mutated genes associated with gliomas

RadElement ID

RDE748

Definition

Identify mutated genes associated with gliomas

Data Type

Multi-select, categorical

Value Set

  • EGFR
  • ERBB2
  • IDH1
  • NF1
  • PIK3CA
  • PIK3R1
  • PTEN
  • PTPRD
  • RB1
  • TP53

Units

N/A



Secondary Outputs


Survival Probability

RadElement ID

RDE741

Definition

Predict survival probability

Data Type

Numeric

Value Set

N/A

Units

Probability of survival for a window of time (month)


Pseudo-progression probability

RadElement ID

RDE742

Definition

probability of pseudo-progression after treatment

Data Type

Numeric

Value Set

0-100%

Units

Probability of pseudo-progression after treatment

Pseudo-response probability

RadElement ID

RDE743

Definition

probability of pseudo-response after treatment

Data Type

Numeric

Value Set

0-100%

Units

Probability of pseudo-response after treatment

Recurrence probability

RadElement ID

RDE744

Definition

probability of recurrence after treatment

Data Type

Numeric

Value Set

0-100%

Units

Probability of recurrence after treatment

Radiation necrosis probability

RadElement ID

RDE745

Definition

Probability of radiation necrosis after treatment

Data Type

Numeric

Value Set

0-100%

Units

probability of radiation necrosis after treatment

Future Development Ideas


Although the main target of the use case is glioma as the most common CNS primary tumor, since the database contains all primary CNS tumors the same prediction can be achieved for other non-glioma tumors (of course if we can collect enough cases).

In addition, in the future, additional gene mutations can be discovered in these tumors. Retraining of the algorithm for future mutations can be possible.

This dataset also can be used to facilitate the clinical trials for treating CNS tumors. Most of the trials work with inclusion criteria and they try to include the most homogeneous patient population into the trial. The same dataset can help the researchers to select more homogeneous patients for their trials.