Deep-learning diagnostics

DiagnoSight AIMedical Diagnostic Assistant

An AI medical assistant using deep learning (TensorFlow) to analyze medical images, generate diagnostic PDF reports, and answer queries via a chatbot.

~/diagnosight/architecture4 layers
  1. 04User Layer
  2. 03Report & Chat Layer
  3. 02Analysis Pipeline
  4. 01AI Core
Condition Categories
15+
Image Types
3
Core Modules
4
Report Generation
PDF

01 / The Problem

DiagnoSight AI speeds up image interpretation by providing instant CNN analysis of medical scans, cross-referencing findings with patient symptoms to output a first clinical opinion and specialist suggestions.

02 / Key Features

  • Instant Medical Image Analysis

    Processes X-rays, MRIs, and CT scans in seconds using a deep learning CNN model built with TensorFlow — returning a confidence-scored prediction across trained condition categories.

  • Diagnosis Tab — 15+ Conditions

    Identifies conditions across 15+ medical categories, providing a quick first opinion on likely diagnosis with probability breakdown and plain-language explanation of findings.

  • AI Support Chatbot

    An in-app conversational chatbot that answers users' follow-up questions about their diagnosis results in plain language — reducing anxiety and helping users understand next steps.

  • PDF Report Generation

    Generates a downloadable, formatted PDF report summarizing the diagnosis, confidence scores, image metadata, and recommended actions — for the user to keep or share with a doctor.

  • Doctor & Specialist Recommendations

    Based on the diagnosed condition, recommends relevant specialist types and helps users locate nearby doctors — ensuring users know exactly where to go for professional care.

  • Full Account System & History

    Complete user authentication, profile editing, password management, and a full saved report history — so users can revisit, compare, or share previous diagnostic sessions.

03 / System Architecture

  1. AI Core

    Deep Learning Model — TensorFlow CNN

    A Convolutional Neural Network trained on labeled medical imaging datasets to classify conditions across 15+ categories. The model is loaded once at startup and performs inference on preprocessed image tensors — returning a probability distribution across all trained classes.

  2. Analysis Pipeline

    Image Preprocessing & Inference Engine

    Accepts JPEG/PNG uploads of X-rays, MRIs, and CT scans. Applies normalization, resizing, and tensor conversion before passing to the CNN. Combines image predictions with symptom inputs to generate a cross-referenced diagnostic result with confidence scores.

  3. Report & Chat Layer

    PDF Generator & NLP Chatbot

    The report generator formats diagnostic results, metadata, and recommended actions into a structured PDF. The chatbot module uses NLP to interpret follow-up questions in natural language and return contextual responses about the diagnosis — keeping users informed without overwhelming them with technical output.

  4. User Layer

    Account System & Report History

    Full authentication — registration, login, profile editing, password change. Each user has a persistent diagnostic history stored in the database, with the ability to view, re-download, or delete past reports. Doctor recommendations are surfaced based on diagnosed condition type and location input.

04 / Tech Stack

  • Python
  • TensorFlow 2.x
  • Keras
  • Convolutional Neural Network
  • NumPy & Pillow
  • NLP Chatbot
  • PDF Generation (ReportLab)
  • User Auth & Database

Explore the Full Codebase on GitHub

The full source — TensorFlow model, inference pipeline, chatbot, PDF generator, auth system, and UI — is available on GitHub.