For employers

Hire this AI Trainer

Sign in or create an account to invite AI Trainers to your job.

LoginCreate account
Invite to Job
Taha Ismail

Taha Ismail

Radiologist with AI research; expert in medical image annotation

India flagBangalore, India
$33.00/hrExpertCVATLabelboxGoogle Cloud Vertex AI

Key Skills

Software

CVATCVAT
LabelboxLabelbox
Google Cloud Vertex AIGoogle Cloud Vertex AI
Other

Top Subject Matter

No subject matter listed

Top Data Types

Geospatial Tiled ImageryGeospatial Tiled Imagery
ImageImage
Medical DicomMedical Dicom

Top Task Types

Bounding Box
Classification
Segmentation

Freelancer Overview

As a radiologist with over five years of clinical and academic experience, I have specialized in annotating and labeling multimodal medical imaging datasets—including CT, MRI, and ultrasound—for use in AI training and research. My work has contributed directly to the development and validation of deep learning models for tumor detection, segmentation, and classification, with a strong focus on ovarian cancer. I have co-authored multiple peer-reviewed research articles exploring advanced techniques such as Inception-ResNet classifiers, quantum convolutional neural networks, fuzzy clustering, and conditional GANs for image-to-image translation. These projects involved extensive pixel-level segmentation, ROI annotation, and clinical labeling to generate high-quality supervised datasets. What sets me apart is my dual expertise in radiological diagnosis and AI-centric data preparation. I have hands-on experience with semantic segmentation, image-level classification labeling, and the integration of clinical metadata with imaging for machine learning pipelines. My ability to bridge clinical insight with technical annotation protocols ensures that the labeled data is both anatomically accurate and model-ready—making me a valuable contributor in AI model training, especially within medical imaging and healthcare applications.

ExpertUrduHindiArabicEnglishKannada

Labeling Experience

CT Image Labeling for Ovarian Tumor Classification and Explainable AI Feature Analysis

OtherTextClassificationDiagnosis
Manually curated and labeled 1,500 ovarian tumor cases from clinical records, including 28 features across imaging, symptoms, and tumor markers. Used classification labeling for benign/borderline/malignant categories. Applied quality control through statistical tests (p-values, Cohen’s d) and explainability checks using SHAP and LIME. Project ensured high model accuracy (98.66%) and reproducibility with clinical relevance.

Manually curated and labeled 1,500 ovarian tumor cases from clinical records, including 28 features across imaging, symptoms, and tumor markers. Used classification labeling for benign/borderline/malignant categories. Applied quality control through statistical tests (p-values, Cohen’s d) and explainability checks using SHAP and LIME. Project ensured high model accuracy (98.66%) and reproducibility with clinical relevance.

2024 - 2024

Ovarian Tumor Detection using Inverted Fuzzy C-Means and Quantum CNN on CT Images

OtherMedical DicomSegmentationClassification
Segmented and annotated over 5,400 CT images of ovarian tumors from 78 patients, including axial, sagittal, and coronal views. Tumors were labeled as benign or malignant using inverted fuzzy C-means clustering with GenClust++ optimization. QCNN and CNN-based classification models (ResNet16, VGG16, Inception-v4) were trained and validated. Quality was ensured via Jaccard/Dice scores (avg 0.77/0.79), classification accuracy (up to 87.02%), and ethical adherence with anonymized datasets.

Segmented and annotated over 5,400 CT images of ovarian tumors from 78 patients, including axial, sagittal, and coronal views. Tumors were labeled as benign or malignant using inverted fuzzy C-means clustering with GenClust++ optimization. QCNN and CNN-based classification models (ResNet16, VGG16, Inception-v4) were trained and validated. Quality was ensured via Jaccard/Dice scores (avg 0.77/0.79), classification accuracy (up to 87.02%), and ethical adherence with anonymized datasets.

2021 - 2023

Comparative Annotation for Ovarian Cancer Segmentation using Active Contour, Random Walker, and Watershed Algorithms

OtherMedical DicomSegmentationDiagnosis
Labeled and segmented ovarian tumor images using three classical computer vision algorithms: Active Contour, Random Walker, and Watershed. Created ground truth segmentation maps based on my (radiologist-defined) ROIs. Project involved comparative validation of algorithmic outputs against expert-annotated masks to identify the most accurate segmentation approach. Ensured consistency in tumor margins and controlled image preprocessing for optimal model evaluation. Result used for benchmarking segmentation accuracy and reproducibility.

Labeled and segmented ovarian tumor images using three classical computer vision algorithms: Active Contour, Random Walker, and Watershed. Created ground truth segmentation maps based on my (radiologist-defined) ROIs. Project involved comparative validation of algorithmic outputs against expert-annotated masks to identify the most accurate segmentation approach. Ensured consistency in tumor margins and controlled image preprocessing for optimal model evaluation. Result used for benchmarking segmentation accuracy and reproducibility.

2022 - 2022

Ovarian Cancer Segmentation and Classification Using Conditional GANs

OtherMedical DicomSegmentationClassification
Annotated and segmented MRI images of ovarian tumors (benign and malignant) with radiologist input for developing a Conditional GAN (cGAN)-based image-to-image translation pipeline. Images were labeled for tumor boundaries and classified by malignancy. The model achieved segmentation scores of 0.825 (benign) and 0.765 (malignant), and classification accuracy of 83% and 79% respectively. Compared model output to expert radiologist interpretations and alternative architectures (UNet + ResNet101) for validation.

Annotated and segmented MRI images of ovarian tumors (benign and malignant) with radiologist input for developing a Conditional GAN (cGAN)-based image-to-image translation pipeline. Images were labeled for tumor boundaries and classified by malignancy. The model achieved segmentation scores of 0.825 (benign) and 0.765 (malignant), and classification accuracy of 83% and 79% respectively. Compared model output to expert radiologist interpretations and alternative architectures (UNet + ResNet101) for validation.

2022 - 2022

Image Annotation for Ovarian Tumor Classification Using Inception-ResNet and QCNN

OtherMedical DicomBounding BoxClassification
The project aimed to classify ovarian tumors as benign or malignant using Inception-ResNet v2 and Quantum CNN models trained on CT scan data. I labeled over 900 CT slices by manually identifying and cropping lesions, drawing bounding boxes, and assigning classification labels. The dataset was organized and balanced for optimal model training. Quality measures included my (radiologist-reviewed) annotations, inter-slice consistency checks, and class distribution validation. The labeled dataset enabled the models to reach up to 97.5% validation accuracy.

The project aimed to classify ovarian tumors as benign or malignant using Inception-ResNet v2 and Quantum CNN models trained on CT scan data. I labeled over 900 CT slices by manually identifying and cropping lesions, drawing bounding boxes, and assigning classification labels. The dataset was organized and balanced for optimal model training. Quality measures included my (radiologist-reviewed) annotations, inter-slice consistency checks, and class distribution validation. The labeled dataset enabled the models to reach up to 97.5% validation accuracy.

2022 - 2022

Education

I

ICRI, IRIA

Masters, Clinical Radiology

Masters
2020 - 2025
N

National Board of Examinations

DNB, Radiodiagnosis

DNB
2024 - 2024

Work History

K

Kanachur Institute of Medical Sciences

Assistant Professor

Mangalore
2025 - Present
T

The Radiology Group Teleradiology

Consultant Radiology Pre-reader

Mangalore
2024 - Present